By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Please be advised that we experienced an unexpected issue that occurred on Saturday and Sunday January 20th and 21st that caused the site to be down for an extended period of time and affected the ability of users to access content on Wiley Online Library. This issue has now been fully resolved. We apologize for any inconvenience this may have caused and are working to ensure that we can alert you immediately of any unplanned periods of downtime or disruption in the future.
Oliver Schweiger, UFZ, Centre for Environmental Research Leipzig-Halle, Department of Community Ecology, Theodor-Lieser-Str. 4, D-06210 Halle, Germany (fax +49 34555 85329; e-mail email@example.com).
1In landscapes influenced by anthropogenic activities, such as intensive agriculture, knowledge of the relative importance and interaction of environmental factors on the composition and function of local communities across a range of spatial scales is important for maintaining biodiversity.
2We analysed five arthropod taxa covering a broad range of functional aspects (wild bees, true bugs, carabid beetles, hoverflies and spiders) in 24 landscapes (4 × 4 km) across seven European countries along gradients of both land-use intensity and landscape structure. Species–environment relationships were examined in a hierarchical design of four main sets of environmental factors (country, land-use intensity, landscape structure, local habitat properties) that covered three spatial scales (region, landscape, local) by means of hierarchical variability partitioning using partial canonical correspondence analyses.
3Local community composition and the distribution of body size classes and trophic guilds were most affected by regional processes, which highly confounded landscape and local factors. After correcting for regional effects, factors at the landscape scale dominated over local habitat factors. Land-use intensity explained most of the variability in species data, whereas landscape characteristics (especially connectivity) accounted for most of the variability in body size and trophic guilds.
4Synthesis and applications. Our results suggest that management effort should be focused on land-use intensity and habitat connectivity in order to enhance diversity in agricultural landscapes. Since these factors are largely independent, specific conservation programmes may be developed with regards to socio-economic and agri-environmental requirements. Changes in either of these factors will enhance diversity but will also result in specific effects on local communities related to dispersal ability and the resource use of species.
In the 20th century agricultural intensification and changes in landscape structure emerged as a serious threat to biodiversity (Robinson & Sutherland 2002). An increasing specialization, often leading to monocultures, fertilization, irrigation and pesticide use as well as the fragmentation and loss of semi-natural habitats, substantially alter the patterns of resource availability and biotic interactions in local communities (cf. Matson et al. 1997). The relationship between environmental factors and local communities, however, differs between organizational levels and spatial scales (Cushman & McGarigal 2002; Steffan-Dewenter et al. 2002; Willis & Whittaker 2002; Grand & Cushman 2003).
Results from the recently completed European Union (EU) research project Greenveins demonstrated robust relationships between species richness of plants, birds and arthropods and both landscape structure and agricultural land-use intensity that apply across temperate Europe (R. Billeter, unpublished data). In order to analyse these relationships in detail, the present study aimed to quantify the relative impact of these environmental factors on local arthropod communities across three organizational levels (taxon, size class, trophic guild). Arthropod species make up about 65% of all organisms (Groombridge 1992), represent good correlates for species diversity (Duelli & Obrist 1998) and have a significant impact on ecosystem processes.
Species–environment relationships were examined in a hierarchical design of four main sets of environmental variables that covered three spatial scales. At the regional scale, we accounted for the effects of regional species pools and other historical influences. At the landscape scale, we analysed several aspects of land-use intensity on arable fields and the landscape structure of semi-natural elements. At the local scale, we evaluated aspects of local habitat properties.
We addressed the following questions. (i) What is the relative impact of scale compared with the effects of environmental factors on local arthropod community composition and structure? (ii) What is the relative influence of land-use intensity, landscape structure and habitat properties on local arthropod community composition? (iii) How are body size and trophic position affected by these factors?
Materials and methods
study sites and environmental variables
Five arthropod taxa, wild bees (Apoidea), true bugs (Heteroptera), carabid beetles (Carabidae), hoverflies (Syrphidae) and spiders (Araneae), were sampled at 24 test sites of 4 × 4 km in agricultural landscapes. The test sites were distributed over seven European countries (see Appendix S1): France (three tests sites), Belgium (four), the Netherlands (four), Switzerland (three), Germany (four), Czech Republic (two), and Estonia (four). Together, these covered gradients of both agricultural land-use intensity and landscape structure.
We recorded environmental variables in a nested, hierarchical design of four main sets and eight subsets that covered three spatial scales (Table 1). At the regional scale, we considered ‘country’ as a main set to account for the biogeographical gradient covering our study and the possible influences of country-specific historical or cultural effects on regional species pools.
Table 1. Nested structure of environmental variables
Seven dummy variables
Landscape (4 × 4 km)
Number of pesticide applications per year
Nitrogen fertilizer applied per hectare and year
Intensely fertilized area
Proportion of green veining
Area weighted proximity index of green veining
Number of habitat types
Local (50 m)
Proportion of green veining
Number of habitat types
At the landscape scale, two main sets were distinguished. First, average land-use intensity (LUI) was evaluated per test site by standardized interviews with farmers about management practices on arable fields (Herzog et al. 2005); we recorded two subsets of LUI. The first consisted of two variables representing environmental stress factors (stress): (i) number of pesticide applications to major crops per year and (ii) amount of nitrogen fertilizer applied per hectare and year. The second subset represented spatiotemporal patterns (spatial): (iii) crop diversity (number of crops in rotation) and (iv) proportion of intensely fertilized area (> 150 kg N ha−1 year−1).
Secondly, landscape structure was focused on in terms of semi-natural elements (e.g. woodlands, hedgerows, ditches, grasslands), so called ‘green veining’ (GV) at the landscape scale. Environmental variables were evaluated from digitized habitat maps using orthorectified aerial photographs with spatial resolutions greater than 1 × 1 m and ArcGIS software (ESRI 2003). The classification of the habitats was based on the European Nature Information System (EUNIS) habitat classification (available at http://mrw.wallonie.be/dgrne/sibw/EUNIS/home.html). We aggregated the semi-natural habitats as GV to contrast the highly dynamic anthropogenic elements of arable land and built-up areas.
To describe the structure of GV at the landscape scale and to separate the impacts of species–area relationships, landscape connectivity and landscape diversity, we recorded three subsets of landscape variables per test site: (i) landscape composition was calculated as the proportion of GV (composition); (ii) landscape configuration was assessed by several landscape structure metrics retrieved from fragstats analyses based on the aggregated GV habitats (‘configuration’; McGarigal et al. 2002); (iii) landscape diversity was measured as the number of GV habitat types based on eunis classification (diversity). As an equal number of variables within each environmental subset is required for reliable comparisons (Grand & Cushman 2003), the landscape configuration metrics retrieved from fragstats (number of patches, edge density, proximity index, Euclidean nearest neighbour distance, patch cohesion index, splitting index, clumpiness index) were subjected to a forward selection procedure in a canonical correspondence analysis (CCA), with all arthropod species and the species of single taxa as the dependent matrices. Although the results differed somewhat between the five arthropod taxa and several metrics accounted significantly for the explanation of community variability, we selected only one configuration metric to retain consistency with landscape composition and landscape diversity. The area weighted mean proximity index of the aggregated GV elements best explained the pooled arthropod data as well as most of the single taxonomic groups, and was significant in all cases. Therefore, this metric was used for further analyses. The proximity index considers both local patch size and distance and is interpreted as a measure of connectivity.
The fourth main set of environmental variables described habitat properties of GV at the local scale. The variables were calculated for circular areas with a radius of 50 m around the arthropod sampling points. We recorded two subsets of local variables: (i) local habitat composition included the proportions of single and aggregated GV habitat types (composition) and (ii) local habitat diversity was measured as the number of GV habitat types (diversity). The local habitat composition variables were subjected to a forward selection procedure in a CCA and resulted in the selection of the proportion of aggregated GV elements for further analyses. This led to a total of 16 explanatory variables, nested within eight subsets, which were in turn nested within four main sets (Table 1). The distribution of these variables across the seven countries is given in Appendix S2.
Spiders and carabid beetles were captured with pitfall traps, whereas wild bees, true bugs and hoverflies were sampled with combined flight traps (a combination of window and yellow pan traps; Duelli, Obrist & Schmatz 1999). The test sites were divided into 16 grid cells of 1 km2. In every grid cell a trap set was placed at a randomly chosen ecotone between a GV habitat and agricultural field. Each trap set consisted of two trap units, which were spaced between 25 m and 50 m apart from each other. A trap unit comprised one pitfall and one combined flight trap. Thus, 16 trap sets comprised 32 pitfall traps and 32 combined flight traps per test site. The circular areas for local scale analysis were created around the centre point between both trap units of a trap set. Sampling was carried out according to Duelli (1997). In order to minimize the sampling effort while maximizing efficiency, we restricted the sampling to two periods of maximum activity and density of the species (7 weeks in autumn 2001 and 5 weeks in early summer 2002). To correct for climatic and consequently phenological differences between the countries, we used the blossoming of Taraxacum officinale Wiggers as a guide to commence sampling. The traps were emptied weekly. To account for differences in weather conditions between the test sites during the sampling periods, we considered only the samples with most specimens for the analysis (4 weeks from the autumn period, 3 weeks from the early summer period). The samples were pooled for each trap set and the specimens were identified to species level.
classification of arthropods
The analysis was performed on different sets of taxonomic and ecological groups. First, we analysed community composition based on the log abundance data from all arthropod species. We removed species captured in less than five trap sets from the data set, resulting in a total number of 628 species represented by 170 105 individuals (Table 2). This analysis was repeated for each of the five arthropod taxa separately. Secondly, we classified species into ecological groups according to body size and trophic guild. In order to account for general differences in size between the arthropod taxa, we created four size classes for each of the five taxa (Table 2).
Table 2. Species and individual numbers within size classes of the five arthropod taxa. Body size was obtained from the literature (references and specific body sizes will be provided by the authors on request)
Additionally, the arthropod species were assigned to four trophic guilds: omnivores, herbivores, predators and saprophages. Contributions to the single trophic guilds differed between the taxa (Table 3). Wild bees, true bugs, carabid beetles and spiders were classified according to the feeding habits of the adults, whereas hoverflies were classified by reference to their larvae. The ecological information was obtained from the literature (references will be provided by the authors on request).
Table 3. Species and individual numbers within trophic guilds. Feeding habits were obtained from the literature (references and classifications will be provided by the authors on request)
hierarchical partitioning of community variation
To quantify the species–environment relationships in detail we hierarchically partitioned the variability in the community data that was explained by specific sets of environmental data (Cushman & McGarigal 2002; Jeanneret, Schupbach & Luka 2003) for the arthropod community as a whole based on either taxonomic (species) or ecological units (size classes, trophic guilds), and replicated this analysis for each of the five arthropod taxa by means of a partial CCA (Borcard, Legendre & Drapeau 1992; Legendre 1998). We used the same set of explanatory variables for all multivariate analyses to allow reliable comparisons between taxonomic and ecological groups (Okland 1999). This also allowed the comparison of environmental factors across organizational levels as well as spatial scales through a series of partial CCA calculating the ‘marginal’ and ‘conditional’ effects of the environmental main sets and subsets. Marginal effects denoted the variability explained by a given set of environmental variables without considering other environmental factors, whereas conditional effects denoted the variability explained by a given environmental set after removing the confounding effect of one or more other environmental variables (covariables). Marginal and conditional effects were tested for significance with Monte Carlo permutation tests by 999 permutations within the particular hierarchical level.
Because of differences in the lack-of-fit of data to the response models for different multivariate data sets, the unexplained variation and hence the ‘total variability explained’ by all environmental factors is generally inappropriate for comparisons of different data sets (Okland 1999). To overcome this problem, we followed Okland (1999) and focused on the explainable variability only, using the ‘proportions of total variability explained’ by the particular sets of environmental variables instead. Hence, we analysed the relative importance of single sets of environmental variables not their absolute impact.
Differences between the proportions of variability explained by the sets of environmental variables were tested for significance by replications over the single taxa by means of a one-way anova. To illustrate the relationships between environmental variables and species or ecological groups at a specific hierarchical level we produced biplots of the respective partial CCA. The analysis was performed with canoco (Ter Braak & Smilauer 2002) and the statistical software package R (R Development Core Team 2004).
The total variability explained by the set of 16 independent variables (country and environment) was 28%, of which the environmental factors accounted for 60% and country-specific characteristics explained 78%, with an overlap and consequent redundancy of 38% in the explanation by country and environment. Hence we focused on the conditional effect of the environment (with country as covariable), which accounted for 22% of the total variability explained.
Environmental factors at the landscape scale explained more variation than local-scale factors (Fig. 1). The main set of LUI variables explained most of the variability, with spatial factors dominating slightly over stress factors. The effect of GV was clearly dominated by configurational aspects (i.e. mean proximity index of GV elements). Local habitat factors were of minor importance. All marginal and conditional effects were significant (P < 0·05). The low level of confounding between the main sets and between the GV and LUI subsets indicated their relative independence. In contrast, the confounding between the local habitat subsets composition and diversity was high (27% of variability explained by composition).
The biplot of the partial CCA (data not shown) including all conditional environmental variables (country as covariable) revealed two major gradients. The first canonical axis explained 20% of the species variation and reflected a GV–LUI gradient from high connectivity (i.e. proximity index; correlation coefficient −0·72) and proportion of GV (correlation coefficient −0·65) to high land-use intensity (correlation coefficients between 0·46 and 0·54). The second axis explained 17% of the species data and represented a pure GV gradient from high connectivity (i.e. proximity index; correlation coefficient −0·43) to high landscape diversity (i.e. number of habitat types; correlation coefficient 0·31).
In separate analyses of the five arthropod taxa, the total variability explained by all independent variables was rather similar (minimum 22% for bugs to maximum 31% for carabid beetles). After removing the confounding effect of country, the environmental variables accounted for a minimum of 20% (carabid beetles) to a maximum of 25% (hoverflies) of the total variability explained.
Despite the fundamental ecological differences between the five arthropod taxa, they were quite similar in their dependence on the environmental factors and confirmed the general trend of the pooled arthropod community (Fig. 2). At the level of the main sets of environmental variables, the observed differences in the effects of LUI, GV and local habitat factors were significant (P < 0·001, anova; Fig. 2a). Post-hoc tests indicated that the proportion of total variability explained by LUI was greater than that of GV and that the effect of GV was greater than that of local habitat features. Visual inspections of Fig. 2a revealed no substantial differences between the arthropod taxa.
Analyses of the LUI subsets at the landscape scale confirmed the dominance of spatial factors over stress factors (P = 0·022, anova; Fig. 2b) and that the strength of the effect of LUI depended on the taxonomic group. Spatial as well as stress factors contributed most to the explanation of variability in bug communities, and least to hoverfly and carabid communities.
Within the main set of GV variables at the landscape scale, the response of the five arthropod taxa was comparatively heterogeneous (Fig. 2c). The proportions of variability explained by the three subsets were significantly different (P = 0·048, anova) but post-hoc tests revealed that the effect of landscape configuration was only marginally significantly higher than that of landscape diversity or composition (both P= 0·076). Visual inspections of Fig. 2c indicated taxon-specific response patterns. For wild bees, spiders and true bugs total variability was explained mainly by configuration followed by diversity and finally composition, whereas for carabid beetles and hoverflies composition was more important than diversity.
Analyses of the habitat-specific subsets at the local scale confirmed the dominance of local habitat composition over local habitat diversity (P < 0·001, anova; Fig. 2d). Their relative effect did not differ between the taxa.
The total variability explained by the independent variables (country and environment) was smaller for size classes (26%) than for trophic guilds (46%) for the pooled data. However, after removing the confounding effect of country, the relative contribution was reversed (39% vs. 18%, respectively). There were pronounced differences in the dependence of size classes and trophic guilds on particular sets of environmental variables (Fig. 3).
The distribution of size classes was most influenced by GV at the landscape scale (Fig. 3a), with landscape configuration being the most important determinant of the local size structure of the arthropod communities. Within LUI, stress factors dominated. The variability explained by local habitat features was not significant (Monte Carlo permutation test).
The biplot including all conditional environmental variables (country as covariable) revealed a strong dependency of the size classes on a landscape configuration gradient (Fig. 4). The first canonical axis explained 63% of the variation and ranged from high to low connectivity (i.e. proximity index; correlation coefficient −0·29), where the size of the arthropods increased with decreasing connectivity. The second canonical axis explained 30% and reflected a pure LUI gradient (e.g. correlation coefficient of pesticides −0·18), where medium-sized arthropods were associated with low levels of LUI and small and large arthropods with high levels of LUI. Large arthropods were particularly associated with frequent pesticide applications.
The distribution of trophic guilds was more evenly affected by the main sets of environmental variables compared with the analysis of the size classes (Fig. 3b). GV accounted for the greatest proportion of variability of trophic guilds but was less dominant than for size classes. Within GV, landscape configuration was most important but was confounded to a great extent by compositional aspects. The influence of diversity at the landscape scale was not significant (Monte Carlo permutation test). Within LUI, spatial factors dominated over stress factors. The influence of local habitat features on trophic guilds was relatively high compared with analyses of size classes and arthropod species. However, the local habitat variables were not entirely independent from GV variables at the landscape scale. Within the local habitat subset diversity dominated but was highly confounded by composition (68% of variability explained by composition). This resulted in a non-significant conditional effect of local habitat composition.
The biplot including all conditional environmental variables (country as covariable) revealed two major gradients (Fig. 5). The first canonical axis explained 58% of the variation in the trophic guilds and represented mainly a landscape configuration gradient from high to low connectivity (i.e. proximity index; correlation coefficient −0·25). The second canonical axis explained 38% and reflected predominantly a LUI gradient that was dominated by crop diversity (correlation coefficient of crop diversity −0·23). Saprophagous arthropods were associated with a high level of connectivity, whereas omnivorous arthropods were associated with a high level of LUI. Predators and phytophages reacted similarly and were associated with high landscape diversity.
The results suggest that local arthropod community composition and the distribution of body size classes and trophic guilds were most affected by regional processes, which highly confounded landscape and local factors. Thus, regional factors and consequently regional species pool effects are crucial for local community composition and must not be ignored (Ricklefs 1987). Moreover, this highlights the importance of large-scale factors in determining the regional species pool, especially as some of these factors, such as climate and speciation, are currently altered by human activities (Chapin et al. 2000; Templeton et al. 2001).
After removing the regional effects (and hence that of the regional species pool), variables describing land-use intensity and the structure of GV were more important at the landscape scale than variables describing GV at the local scale. One explanation is that landscape properties will affect local recruitment of species from the regional pool as well as their persistence or local extinction (references in Lawton 1999), whereas local habitat properties might support local persistence just as long as landscape features allow for viable spatial population dynamics (Hanski & Gilpin 1997).
This suggests a positive relationship between scale and the impact of environmental factors on local communities, and supports a top-down hierarchical structure of environmental filters determining local community assembly (Whittaker, Willis & Field 2001; Noda 2004).
Notwithstanding the large geographical range and the associated large regional effect, we found substantial impact of environmental variables on local arthropod communities at the landscape and local scale. The total variability in community data explained by the environmental factors (arthropod taxa 22–31%, size classes 26%, trophic guilds 46%) was well within the usual range one can achieve with a CCA (Okland & Eilertsen 1994; Githaiga-Mwicigi, Fairbanks & Midgley 2002). This highlights the importance and general operation of landscape structure, land-use intensity and habitat diversity in agricultural landscapes.
Previous analysis of the same data set revealed that arthropod species richness at the landscape scale as well as at the local scale decreased with increasing land-use intensity (i.e. spatial components) and decreasing landscape structure (i.e. area and connectivity; authors’ unpublished data). The current analysis of arthropod community composition aimed to disentangle and quantify the relative impact of these environmental factors. Our results confirmed that the arthropod communities reacted predominantly to an intensification gradient from landscapes with a high proportion and connectivity of semi-natural elements to landscapes with high levels of land-use intensity. The low level of confounding between land-use intensity, GV and local habitat properties indicates their relative independence in affecting local community composition. Therefore, changes in either one of these three environmental sets induce specific changes in local arthropod communities and thus in biodiversity, providing great scope for conservation management. As local habitat properties were shown to be less important, management activities may focus on both landscape structure and land-use intensity in order to restore a maximum of diversity. This means that even in areas where agricultural land use has to remain intensive, biodiversity may be enhanced by modifying the landscape structure. On the other hand, in areas where the landscape structure is already rich the focus may be on decreasing land-use intensity rather than on the local management of semi-natural habitats.
The particular importance of agricultural management practices on local arthropod community composition has been demonstrated at smaller spatial scales (Ostman et al. 2001; Jeanneret, Schupbach & Luka 2003). In the present study, however, we were able to show that the spatial and stress-related components of agricultural land use independently affect local arthropod communities to a similar degree. This again provides the possibility of targeting conservation management action when aiming at decreased land-use intensity (either reduce fertilizer and pesticides or increase crop diversity and the proportion of extensively managed fields).
When diversity is managed via landscape structure, our results suggest focusing on connectivity. Landscape composition (i.e. amount of semi-natural habitat), diversity and configuration (i.e. connectivity) have previously been shown to be key factors in determining local communities (Miller, Brooks & Croonquist 1997; Di Giulio, Edwards, & Meister 2001). Habitat loss, rather than fragmentation per se, has been reported to negatively affect biodiversity (Fahrig 2003); however, we found fragmentation per se to be the most important landscape factor affecting local arthropod communities. These findings were confirmed by the surprisingly similar reactions of all taxa despite their obvious ecological differences, indicating some underlying general processes.
Our studies suggest that generic processes of local community assembly operate on species-specific ecological characteristics, quite independently of the taxonomical status, that are reflected by the distributions of body sizes and trophic guilds. There is an ongoing debate as to whether ‘niche assembly’ or ‘dispersal assembly’ drives local community composition (Hubbell 2001; Williams, Jones & Hartley 2001; Whitfield 2002). Niche–assembly theories posit that resource distribution and biological interactions are responsible for species’ coexistence (Williams, Jones, & Hartley 2001). In contrast, neutral dispersal–assembly theories hypothesize that chance, history and dispersal explain the community structure (Bell 2001; Hubbell 2001). Our results suggest interdependency between niche– and dispersal–assembly.
Analysis of body sizes and trophic guilds revealed a dominant impact of the landscape structure. The spatial connectivity of semi-natural habitats predominantly determined the distribution of body sizes and trophic guilds. The relationship between landscape configuration and spatial population dynamics (Hanski & Gilpin 1997), and therefore the significance of dispersal limitation, is supported by the observed ordination of size classes along the connectivity gradient. The size of arthropods increased with decreasing connectivity, indicating that small species cope better with connected than isolated habitats and large species cope better with isolation than small species. Assuming that dispersal ability is positively correlated with body size (Peters 1986; Gathmann & Tscharntke 2002), these findings emphasize that arthropod communities within agricultural landscapes are predominantly affected by dispersal limitation.
Despite the similarity in the general patterns, ecological differences between the arthropod taxa were reflected in the somewhat heterogeneous response to landscape configuration, diversity and composition. Configuration was the most important factor for four of the arthropod taxa but its effect on hoverflies was low. In contrast, hoverflies were affected more strongly than any other taxon by landscape composition. This might be a consequence of the high mobility of this taxon compared with the others. Highly vagile species without significant dispersal limitation are affected by habitat loss only and not by connectivity per se (Tscharntke & Brandl 2004). Hence, conservation management for species or species groups with low dispersal ability should focus on increasing connectivity, while species or species groups with fairly high dispersal ability seem to be supported best by increasing the share of (semi)natural habitats (Chacoff & Aizen 2005).
Our results further suggest that changes in agricultural land-use intensity affect arthropod communities through niche breath and dispersal abilities as well. Small and large arthropods were associated with higher levels of land-use intensity (i.e. stress factors) than medium-sized arthropods. This may be a consequence of both pesticide tolerance and high recolonization ability according to dispersal ability or reproduction. Large arthropods were highly dominated by carabid beetles, some of which appear to be comparatively tolerant to pesticide application (Holland, Winder & Perry 2000), whereas small-bodied species may take advantage of fast reproduction rates after population declines caused by frequent disturbances (Peters 1986).
The ordination of trophic guilds reflected different levels of resource-dependence and spatiotemporal resource variation. The association of omnivorous species with high levels of LUI was most probably the result of their larger trophic niche breadth and hence greater resilience to frequent reduction in food supply as a result of agricultural management. Predators and phytophages reacted similarly in their association with low levels of land-use intensity, probably indicating trophic links.
These results confirm a stimulating effect of low land-use intensity on phytophages as well as predators, indicating possibilities of natural pest control. High abundances and species numbers of predators and alternative prey (in semi-natural habitats) may ensure the presence of a (specialized) predator when pest densities are about to rise (Bianchi & van der Werf 2004). This means that enhancing diversity through changing species composition by lowering land-use intensity may favour natural pest regulation as well.
Hierarchical variability partitioning by means of partial CCA demonstrated high analytical power in unravelling the relative effects of regional factors, aspects of land-use intensity, components of landscape structure and local habitat properties on local arthropod community composition. Our results suggest focusing on two environmental variable sets in order to enhance diversity in European agricultural landscapes: land-use intensity and habitat connectivity. Their independence provides scope for specific conservation programmes with regards to socio-economic and agri-environmental requirements. Changes in either one of these factors will enhance diversity but will also result in specific effects on local communities. Improving connectivity might be favoured as dispersal limitation seems to be of particular importance for local community assembly. This supports agri-environmental programmes that usually try to implement a higher share and connectivity of semi-natural habitats (Sepp et al. 2004). Decreasing spatial or stress factors of land-use intensity, on the other hand, seems to facilitate more complex food webs and thus may favour natural pest regulation. In this context, agri-environmental programmes that aim at the protection of water and therefore try to reduce land-use intensity can be advantageous for terrestrial arthropod communities, although they are not targeted for biodiversity conservation per se.
Many thanks to the taxonomic specialists Tim Adriaens, Frank Burger, Rafaël De Cock and Jaan Luig (bees), Roland Bartels, Jean-Yves Baugnée and Ralph Heckman (bugs), Konjev Desender, Ringo Dietze, Rein Karulaas, Keaty Maes and Viki Vandomme (carabid beetles), Martin Musche and Dieter Doczkal (hover flies), Herman De Koninck, Mart Meriste, Johan Van Keer and Valerie Vanloo (spiders) and many field and laboratory assistants. We are grateful to Roland Brandl and Ingolf Kühn for discussion and comments on statistics and ecology. Funding was received from the Energy, Environment and Sustainable development Programme (FP5) of the European Commission (contract number EVK2-CT-2000-00082).