How landscape structure, land-use intensity and habitat diversity affect components of total arthropod diversity in agricultural landscapes


Frederik Hendrickx, Terrestrial Ecology Unit (TEREC), Department of Biology, Ghent University, K.L. Ledeganckstraat 35, 9000 Gent, Belgium (fax +31 9264 87 94; e-mail


  • 1Agricultural intensification poses a serious threat to biodiversity as a consequence of increased land-use intensity, decreased landscape heterogeneity and reduced habitat diversity. Although there is interest in the preservation of total species richness of an agricultural landscape (γ diversity), the effects of intensification have been assessed primarily by species richness at a local scale (α diversity). This ignores species richness between local communities (β diversity), which is an important component of total species richness.
  • 2In this study, measures of land-use intensity, landscape structure and habitat diversity were related to γ, α and β diversity of wild bees (Apoidea), carabid beetles (Carabidae), hoverflies (Syrphidae), true bugs (Heteroptera) and spiders (Araneae) within 16 local communities in 24 temperate European agricultural landscapes.
  • 3The total landscape species richness of all groups was most strongly affected by increased proximity of semi-natural habitat patches. Bees also decreased in landscapes with a high intensity of farmland management, demonstrating additive effects of both factors.
  • 4Separating total species diversity into components, the decrease in total species richness could be attributed primarily to a decrease in species diversity between local communities. Species richness of the local communities of all investigated groups decreased with increasing land-use intensity and, in the case of spiders, decreasing proximity of the semi-natural habitat patches.
  • 5The effect of increased habitat diversity appeared to be of secondary importance to total species richness but caused a shift in the relative contribution of α and β diversity towards the latter.
  • 6Synthesis and applications. This study demonstrates that the effects of agricultural change operate at a landscape level and that examining species diversity at a local level fails to explain the total species richness of an agricultural landscape. The coincidence of patterns of β diversity with those of γ diversity emphasizes that such information is of crucial importance for the implementation and evaluation of restoration programmes aiming to restore sustainable countryside diversity. As local extinction processes in highly fragmented landscapes shape biodiversity, priority should be given to the conservation of diverse agricultural landscape remnants in Europe.


Agricultural landscapes cover the vast majority of non-urbanized areas in Europe. During the long history of agriculture, numerous species have adapted to this widespread form of land use. The persistence of structurally diverse agricultural landscapes is hence a prerequisite for the conservation of a significant part of Europe's biodiversity (Bignal 1998; Krebs et al. 1999). However, land dedicated to food production has changed dramatically during the post-Second World War period. In parallel with an increased input of pesticides and fertilizers, as well as increased livestock densities, an almost irreversible change of the spatial structuring of the agricultural landscape has taken place (Krebs et al. 1999). The originally diverse, highly structured landscapes have been converted into much more uniform areas consisting almost solely of intensively used agricultural units (Stoate et al. 1991; Robinson & Sutherland 2002), which has coincided with a reduction of species diversity, as has been demonstrated in particular for birds (Siriwardena et al. 1998; Krebs et al. 1999; Benton, Vickery & Wilson 2003; Heikkinen et al. 2004).

Previous studies have demonstrated that increased management intensity of the agricultural fields is one of the main causes of the decline of local species richness (Bengtsson, Ahnström & Weibull 2005; Dauber et al. 2005). It was predicted that local species diversity would be enhanced under organic farming management but this has not been widely observed, probably because of the confounding effects of the surrounding landscape (Weibull, Bengtsson & Nohlgren 2000; Kleijn et al. 2001). This demonstrates that ecological processes acting on spatial scales larger than the local environment may additionally impact local species diversity and often have an interactive affect with land-use intensity (Tscharntke et al. 2005). Such landscape effects can be manifested at two non-exclusive levels, landscape and habitat heterogeneity (Weibull, Östman & Granqvist 2003; Tews et al. 2004), as well as the amount of semi-natural habitat (Bergman et al. 2004; Clough et al. 2005; Schmidt et al. 2005). In a previous study, local environmental factors proved to be subordinate to environmental factors at the landscape scale in explaining the total variation in local species composition (Schweiger et al. 2005).

These observations emphasize that the preservation of diverse agricultural landscapes should focus on species enhancement of entire agricultural areas rather than just on diversity of local communities. The decrease in total species diversity of an agricultural landscape can be seen with the species diversity of the local community, i.e. alpha diversity (α), and the diversity between local communities, i.e. beta diversity (β). Both components contribute to gamma diversity (γ) (Whittaker 1972; Lande 1996; Veech et al. 2002), which in this context is the total diversity of an agricultural landscape. Until now, changes in the components of landscape species diversity in relation to agricultural or landscape features remain largely unexplored, although such an approach can yield additional insight into how (environmental) factors affect species diversity (Loreau 2000; Wagner, Wildi & Ewald 2000; Veech et al. 2002).

Agricultural landscapes can be viewed as a mosaic of habitat islands sustaining high levels of biodiversity embedded in a matrix of more intensively used agricultural land (Duelli 1997; Wagner, Wildi & Ewald 2000; Fournier & Loreau 2001). The size as well as spatial configuration of these semi-natural habitat patches can be important determinants for sustaining a (meta) population of less ubiquitous species (Hanski et al. 1995; Hanski 1998). Consequently, when suitable habitat patches become simultaneously reduced in size and more distant from each other, local extinction rates as a result of stochastic intrinsic and extrinsic effects are not compensated for by recolonization if the dispersal capacities of the involved species are low (Thomas 2000; Fahrig 2003; Colas, Thomas & Hanski 2004; Parvinen 2004). This can lead to a strong impoverishment of the local diversity. Despite the convincing theoretical and empirical framework concerning the expected loss of single species in response to habitat loss, studies relating the local community diversity to the amount of suitable habitat still remain scant (but see Robinson et al. 1992; Burkey 1995; Bergman et al. 2004; Summerville & Crist 2004). Also, information on the extent to which similar or dissimilar species become eliminated from the habitat patches or, alternatively, invade from the intensively used agricultural matrix is scarce. However, as only a small set of species is adapted to the high disturbance levels of agricultural fields (Maelfait & De Keer 1990; Samu & Szinetar 2002), uniform landscapes as well as intensive farming systems can lead to a strong homogenization of the local communities, resulting in a decrease in diversity between local communities.

When investigating biodiversity responses, multiple species group studies are most suitable for gaining insights into the effects of intensification of agriculture. The diverse nature of arthropods allows (i) investigation of the responses of species covering a wide range of ecological functions and (ii) identification of the ecological characteristics of the organisms most sensitive to environmental change (Duelli 1997; Duelli, Obrist & Schmatz 1999). Indeed, in their review concerning the effects of post-war changes in arable farming on biodiversity in Britain, Robinson & Sutherland (2002) concluded that data for many taxa are too scarce at present to permit a detailed assessment of the factors involved. This is particularly true if the aim is to separate the effects of increased land-use intensity, altered landscape structure and decreased habitat diversity (but see Schweiger et al. 2005). Although these effects are often associated with each other in practice, separating them is of great importance for designing guidelines to restore a sustainable countryside (Sutherland 2002; Tscharntke et al. 2005), as is the case in agri-environment schemes. Recent evaluations of these schemes reveal that the results are rather equivocal (Kleijn et al. 2001; Kleijn & Sutherland 2003), which might be partly because of the confounding effects of large-scale landscape structure (Sutherland 2002; Kleijn & Sutherland 2003; Bergman et al. 2004; Bengtsson, Ahnström & Weibull 2005; Tscharntke et al. 2005) but additionally because diversity has been primarily explored at the local scale.

Assessing how the different diversity components of landscape species richness react to the major factors related to agricultural change should therefore provide indispensable information for identifying the appropriate scale for the implementation of agri-environment schemes (Kareiva & Wennergren 1995). In this study, we addressed whether (i) the total species richness is affected by increased land-use intensity, homogenized landscape structure and decreased habitat diversity and (ii) this is caused either by a decrease in local species richness (α diversity) and/or an additional decrease in species diversity between local communities (β diversity). These diversity components were estimated for five different groups of arthropods covering different ecological niches and assessed for 16 local communities in 24 agricultural landscapes of temperate Europe.

Materials and methods

study area and sampling design

The landscape test sites (LTS) were spread over seven temperate European countries: Belgium (four), Czech Republic (two), Estonia (four), France (three), Germany (four), the Netherlands (four) and Switzerland (three). Each LTS measured 16 km2 and was divided into 1-km2 grid cells wherein one trap set was located. As we wanted to sample the local community of the more natural habitat patch as well as the agricultural field, trap sets were located at the border of a randomly chosen semi-natural habitat patch and an agricultural field. A patch was considered semi-natural if it was unmanaged or managed to the extent that its natural production levels were not purposely increased. This strategy ensured that the species composition of the samples was not influenced by the distance between the trap set and the agricultural field, which would complicate standardization of the sampling procedure as habitat patches strongly differed in size. However, the geographical standardization of our sampling scheme did not allow for sampling the same habitat type in each LTS or within each 1-km2 grid cell.

A trap set consisted of two pitfall traps (diameter 10 cm; half-filled with formalin solution) and two combined flight traps that were separated between 25 and 50 m from each other. Flight traps consisted of two vertical orthogonal crossed plastic window glasses above a yellow pan trap (Duelli 1997).

Although the spatial scale of the study did not allow for year-round sampling, the study was conducted over 7 weeks in summer–autumn 2001 and 5 weeks in spring 2002. Traps were emptied weekly and only the most abundant samples were taken into consideration (4 and 3 weeks in summer–autumn and spring, respectively). Full bloom of Taraxacum officinalis Wiggers was used as an indicator to standardize the period of sampling in each country and minimize differences in species composition as a result of different phenological properties of the organisms and climatic conditions between the countries (Duelli 1997; Duelli, Obrist & Schmatz 1999).

Five taxonomic groups, i.e. wild bees (Apoidea), true bugs (Heteroptera), carabid beetles (Carabidae), hoverflies (Syrphidae) and spiders (Araneae), were sorted, and adult specimens were identified to species level. For bees, bugs and hoverflies, only specimens from the combined flight traps were considered, and for carabids and spiders only specimens from pitfall traps were taken into consideration.

landscape and land-use intensity measurements

A large set of characteristics in each LTS was assessed and condensed into one measure of land-use intensity, four measures of landscape structure and two measures of habitat diversity (Table 1). An estimate of land-use intensity was obtained by interviewing at least 10 individual farmers per LTS, although only two to four interviews were performed where the area of managed land per farm was extremely large. Questionnaires gathered information on (i) fertilization application on grassland and arable crops, (ii) pesticide use (herbicides, insecticides, fungicides and retardants) and (iii) livestock densities. These indicators of agricultural land use can be considered independent of landscape properties. Information regarding these three main indicators was condensed into one single measurement of overall land-use intensity per LTS, referred to as the LUI (land-use intensity index). Summary results of the questionnaires and details about the calculation of the LUI can be found in Herzog et al. (2006).

Table 1.  Summary of explanatory variables and their abbreviation
Land-use intensity
LUIIndex of land-use intensity of the agricultural fields based on amount of fertilizer application, livestock densities and use of pesticides (for details see Herzog et al. 2006)
Landscape structure of natural and semi-natural elements
PPERCPercentage of the landscape covered with semi-natural habitat patches
PPERCPercentage of the landscape covered with semi-natural habitat patches
PPROXMean proximity index, wherein proximity combines the size of the focal patch with the distance to other patches (for details see Gustafson & Parker 1992)
PSIZEAverage size of the semi-natural habitat patches
PDENSPatch density
Habitat diversity of natural and semi-natural habitat patches
HDIVMeasure of habitat diversity based on the mean Euclidian distance between the habitat composition around the trap sets within a landscape
NEUNISNumber of European Nature Information System (EUNIS) habitat categories present within the landscape

Measures of landscape structure were obtained from digitized habitat maps derived from aerial photographs and based on the amount and configuration of natural and semi-natural elements in the agricultural landscape. Field observations were used to update the photographs, discriminate between the major types of land use (annual crops, grazing) and define the main types of semi-natural elements (field boundaries, woods, rough grasslands, moorland, etc.).

The total percentage of the landscape covered by semi-natural habitat patches (PPERC), the average patch size (PSIZE) and the density of semi-natural patches (PDENS) were calculated for each LTS. To account for spatial configuration of the habitat patches, we also calculated the mean proximity index (PROXM), which is distance weighted and area based because it sums the ratios of patch area to distance for all habitat patches that fall within some specified distance (boundary of the LTS) to the focal patch (Gustafson & Parker 1992). Hence a high proximity index refers to a landscape with large patches situated close to each other.

Habitat diversity, i.e. the range of habitat types within an LTS, was based on the (i) number of different habitats present within the landscape, wherein habitats were categorized according to the European Nature Information System (NEUNIS) habitat classification (, accessed 30 October 2006) and (ii) mean Euclidian distance in habitat composition between the trap sets within an LTS (HDIV). The latter was assessed by estimating the percentage coverage of the different EUNIS habitat types in a circle with radius of 50 m around the trap sets.

diversity estimates

The hierarchical sampling design enabled us to decompose the total biological diversity within an LTS (γ diversity) into α diversity (i.e. diversity within one trap set) and β diversity (i.e. diversity between trap sets within an LTS).

Despite the standardized sampling protocol, variation in the numbers of trapped arthropods was high and might therefore have influenced species richness estimates substantially (Gotelli & Colwell 2001; Magurran 2004). Indeed, a strong and highly significant correlation between the number of captured species and number of captured individuals was observed for each taxonomic group (n = 23, 0·76 < r < 0·98, P < 0·0001). As the number of captured individuals can be influenced by activity patterns (for example as a result of weather conditions), we used abundance-based methods that estimate species richness independent of sample size (Brose & Martinez 2004).

Rarefaction curves allowed for comparison of species number independently of number of individuals captured (Gotelli & Colwell 2001) and enabled estimation of γ diversity and separation into α and β diversity (Olszewski 2004). Also, the number of species for a local sample size was estimated by means of a sample-based rarefaction curve and used as an estimate of average α diversity within an LTS. Gamma diversity was calculated by estimating the number of species for a regional, i.e. landscape, sample size by means of rarefaction curves in which individuals were randomized among trap sets within an LTS. Two measures of β diversity were calculated as (i) the number of species attributed to differences between local communities, which was obtained by taking the difference between α and γ diversity, and (ii) the relative contribution of β diversity to γ diversity. Rarefaction curves were calculated with the software program EstimateS Vs. 6 (Colwell 2004).

Diversity metrics were calculated for a sample size that was equal to the nearest round number to the least abundant sample for each taxonomical group. However, sample size of one Czech LTS fell far below those of the other LTS and was therefore excluded for analysis. The average number of individuals per LTS (landscape scale)/trap set (local scale), respectively, per taxonomic group was n= 963/61 for Apoidea, n= 2236/140 for Araneae, n= 2814/175 for Carabidae, n= 785/51 for Heteroptera and n= 401/26 for Syrphidae.

statistical analysis

Estimates of the diversity components were related to the environmental variables presented in Table 1 by means of general linear mixed models. Preceding these statistical analyses, we explored the correlation structure among the independent variables in detail by means of correlation graphs, simple linear correlation coefficients and principal component analysis (PCA).

A first set of models was based on fitting each independent variable individually to the diversity estimates. We explicitly did not rely on the obtained PCA axes as they obscure additive and interactive effects among variables that are partially correlated. As we also wanted to test whether a heterogeneous response was present between the different taxa, taxonomic group as well as its interaction with the environmental variable was included as a fixed effect. Different intercepts and slopes were estimated for each taxonomic group separately (Neter et al. 1996). To investigate which fixed effect variable was most important in explaining the observed variation in each diversity component, Akaike's information criteria (AIC) was calculated for each model (based on maximum likelihood). This measure of model fit is based on the −2 log likelihood of the model but controls for the number of free parameters (cf. Verbeke & Molenberghs 2000).

Additional factors were then included in the model to test for additive effects and/or interactive effects between the land-use intensity, landscape structure and habitat diversity variables. We used a stepwise selection procedure in which we started with the model that gave the lowest value of AIC in the first step. Thereafter, all remaining variables, including their interaction with taxonomic group, were tested using forward stepwise regression and were retained when significant and their AIC value computed. The model with the lowest AIC value best explained the observed variation in diversity among the investigated variables.

To correct for possible country effects on the diversity measures (i.e. climatic factors, differences in average habitat type) that were not of main interest in this study, country as well as its interaction with taxonomical group was included as a random effect in each model (Verbeke & Molenberghs 2000).

As absolute values of environmental variables differed strongly, they were standardized to have a mean of zero and a standard deviation of one, which eased comparison of their effects based on regression coefficients.


correlation among landscape variables

Correlations among the variables are presented in Table 2. On average, correlations were low, although some significant correlations were present after sequential Bonferroni correction. Land-use intensity was negatively correlated with the proximity index, and the percentage of the landscape covered with semi-natural patches (PPERC) was positively correlated with habitat diversity and mean proximity index.

Table 2.  Correlations (rP) among explanatory variables measured on the 23 landscape test sites (LTS).*Significant correlation at P < 0·05. The last two columns show correlations of the variables with principal component axes 1 and 2
LUI 1      −0·41 0·31
PPERC−0·47*1      0·50 0·28
PPROX−0·64*0·62* 1     0·49−0·23
PSIZE−0·47*0·32 0·47* 1    0·34−0·30
PDENS 0·240·32−0·32−0·261   0·01 0·67
HDIV−0·280·80* 0·58 0·210·321  0·45 0·30
NEUNIS 0·030·36 0·08 0·120·270·211 0·17 0·39

Based on these LTS characteristics, a PCA was conducted to inspect further the correlation structure among the independent variables. The first and second principal component axis explained, respectively, 43·5% and 25·4% of the total variability in LTS characteristics, respectively (Fig. 1). Values of the eigenvectors revealed that LTS with high values along the first principal component axis were characterized by a large amount of semi-natural habitat patches, high habitat diversity and low land-use intensity (Table 2). Although LTS of each country were situated in each other's neighbourhood according to their PCA scores, overlap along the first axis was large and they did not appear as separate clusters. Along the second axis, which mainly comprised the effect patch density, each country's LTS appeared more or less grouped.

Figure 1.

Principal component ordination of the investigated agricultural landscape test sites based on agricultural land-use intensity, four landscape metrics of natural habitat patches (proportion, average proximity, density and average size) and two habitat diversity metrics.

gamma diversity

Relating each environmental variable separately to the estimates of γ diversity revealed that land-use intensity on the one hand, and percentage of the landscape covered with semi-natural patches, proximity of these patches and both measures of habitat diversity on the other hand, significantly affected total landscape species richness in a negative and positive direction, respectively (Table 3). Except for percentage of the landscape covered with patches, responses were not homogeneous across the taxonomic groups: wild bees and, to a lesser extent, spiders, appeared to be most strongly affected. Among these variables, proximity index best explained the observed variation in total species richness of the agricultural landscape.

Table 3.  Effect of land-use intensity, landscape structure and habitat diversity on the different diversity components based on single variable regression analyses allowing the relationships to differ among the taxonomic groups. Asterisks indicate significance of taxonomic group specific regression slopes after Bonferroni correction (*0·05 > P > 0·01; **0·01 > P > 0·001; ***P < 0·001)
Diversity componentVariableMain effectAICOverall slopeInteractionTaxon-specific slope
GammaLUI17·95  0·00062·570·049 757·6 −7·03 ± 1·59*** −4·47 ± 1·59* −1·07 ± 1·59 −1·76 ± 1·59 −1·79 ± 1·59
PPERC10·49  0·002 763·1   2·05 ± 0·631·740·15 763·7
PDENS 0·26  0·6 772·9  −0·44 ± 0·861·060·4 776·0
PSIZE 0·83  0·4 772·4   0·79 ± 0·870·530·7 778·5
PPROX25·49< 0·00012·60·041 750·1  5·98 ± 1·33***  4·62 ± 1·33**  2·75 ± 1·33  0·85 ± 1·33  1·66 ± 1·33
NEUNIS 7  0·012·490·048 765·3  4·01 ± 1·38*  1·26 ± 1·38  3·10 ± 1·38 −1·55 ± 1·38  2·13 ± 1·38
HDIV10·48  0·0023·430·01 758·4  4·55 ± 1·21**  2·92 ± 1·21*  2·02 ± 1·21 −1·42 ± 1·21  1·18 ± 1·21
AlphaLUI11·14  0·003 575·1  −1·15 ± 0·341·280·3 580·0
PPERC 2·89  0·1 2·670·04 581·7  1·25 ± 0·60  0·98 ± 0·60  0·92 ± 0·60 −1·15 ± 0·60  0·36 ± 0·60
PDENS 1·21  0·3 586·6  −0·35 ± 0·361·020·4 589·9
PSIZE 0·08  0·8 587·6   0·11 ± 0·381·180·3 591·3
PPROX10·67  0·001 3·390·01 571·3  1·37 ± 0·62  2·12 ± 0·62**  1·42 ± 0·62 −0·80 ± 0·62  0·39 ± 0·62
NEUNIS 1·73  0·2 3·280·01 582·1  1·48 ± 0·63 −0·06 ± 0·63  1·48 ± 0·63 −1·05 ± 0·63  0·25 ± 0·63
HDIV 1·77  0·2 4·640·002 575·7  1·49 ± 0·55*  0·98 ± 0·55  0·59 ± 0·55 −1·61 ± 0·55*  0·26 ± 0·55
BetaLUI14·1  0·0024·40·005 683·8 −5·53 ± 1·07*** −2·14 ± 1·07  0·28 ± 1·07 −1·95 ± 1·07 −1·35 ± 1·07
PPERC11·7  0·001 691·7   1·63 ± 0·471·340·3 693·8
PDENS 0·1  0·8 702·8  −0·20 ± 0·650·850·5 706·6
PSIZE 1·08  0·3 701·8   0·70 ± 0·660·270·9 708·7
PPROX22·02< 0·00012·80·03 680·3  4·94 ± 0·97***  2·31 ± 0·97*  1·13 ± 0·97  1·94 ± 0·97  1·15 ± 0·97
NEUNIS 5·53  0·02 698·0   1·25 ± 0·531·220·3 700·7
HDIV12·22  0·0008 691·3   1·57 ± 0·451·380·2 693·6
Beta relativeLUI 2·39  0·146·740·0006349·3−0·053 ± 0·012*** 0·003 ± 0·012 0·019 ± 0·012−0·009 ± 0·012−0·016 ± 0·012
PPERC 8·93  0·0042·880·03−340·5 0·035 ± 0·011** 0·002 ± 0·011 0·005 ± 0·011 0·032 ± 0·011* 0·017 ± 0·011
PDENS 0  0·9−327·8−0·0003 ± 0·0070·860·5−324·0
PSIZE 2·41  0·12−330·3  0·011 ± 0·0071·540·2−329·2
PPROX 7·39  0·0094·550·002−345·2 0·048 ± 0·011***−0·004 ± 0·011−0·004 ± 0·011 0·027 ± 0·011* 0·016 ± 0·011
NEUNIS 5·11  0·03−332·7  0·014 ± 0·0061·210·3−330·1
HDIV10·09  0·002−337·5  0·017 ± 0·0051·880·12−337·6

A model with land-use intensity, its interaction with taxonomic group and proximity of the habitat patches best explained variation in γ diversity. The interaction between taxonomic group and land-use intensity was mainly caused by the strong decrease in γ diversity of bees with increasing farming intensity of the agricultural fields (Table 4 and Fig. 2). This model provided statistical evidence that, at least for bees, land-use intensity and landscape features imposed additive effects on γ diversity.

Table 4.  Effect of land-use intensity, landscape structure and habitat diversity on the different diversity measures, based on multiple regression analyses allowing the relationships to differ among the taxonomic groups. Asterisks indicate significance of taxonomic group specific regression slopes after Bonferroni correction (*0·05 > P > 0·01; **0·01 > P > 0·001; ***P < 0·001)
Diversity componentFactord.d.f.FPAICSlopes
GammaTAX 18·265·43< 0·0001 748·3
LUI 27·5 2·90  0·10 
TAX × LUI 53·2 2·65  0·04   −5·18 ± 1·62*−2·93 ± 1·62 0.26 ± 1·62   0·07 ± 1·62 −0·07 ± 1·62
PPROX 50·611·55  0·0013  2·50 ± 0·74**
AlphaTAX 27·948·96< 0·0001 569·1
LUI107 4·29  0·04 −0·78 ± 0·38*
PPROX104 2·54  0·11 
TAX × PPROX101 3·50  0·01   0·99 ± 0·60  1·74 ± 0·60* 1·03 ± 0·60 −1·11 ± 0·60   0·01 ± 0·60
BetaTAX 16·662·41< 0·0001 674·7
LUI 26·4 0·98  0·3 
TAX × LUI 38·3 4·48  0·005  −4·05 ± 1·13**−0·76 ± 1·13 1·71 ± 1·13 −0·39 ± 1·13   0·09 ± 1·13
PPROX 49·211·94  0·001  1·98 ± 0·57***
Beta relativeTAX 16·414·65< 0·0001−355·9  0·49 ± 0·01  0·57 ± 0·01 0·55 ± 0·01   0·60 ± 0·01   0·54 ± 0·01
LUI 30·3 0·19  0·7 
TAX × LUI 31·3 6·88  0·0004 −0·045 ± 0·012** 0·011 ± 0·0120·027 ± 0·012−0·001 ± 0·012−0·008 ± 0·012
HDIV 80·2 9·10  0·003 0·016 ± 0·005
Figure 2.

Gamma diversity of five different arthropod groups in 23 agricultural landscape test sites in relation to agricultural land-use intensity and average proximity index of the semi-natural habitat patches (black circles, Apoidea; white circles, Araneae; black triangles, Carabidae; white triangles, Heteroptera; black squares, Syrphidae). Graphs represent results conditional upon the effect of other variables.

alpha diversity

Single variable regression analyses revealed that α diversity increased significantly with decreasing levels of land-use intensity and increasing levels of average proximity of semi-natural habitats across taxonomic groups. However, the effect size differed among them, and spiders appeared to exhibit the strongest increase in α diversity when the semi-natural habitat patches of the landscape were characterized by a higher proximity (Table 3). The proximity index best explained the observed variation in α diversity among the seven variables investigated. Effects of habitat diversity did not affect diversity similarly across taxonomic groups; while local bee species richness increased with higher levels of habitat diversity, an opposite pattern was observed for bugs.

Land-use intensity and proximity index were selected as the set of variables that best explained the total variation in local species richness, although the positive effect of increased proximity of the patches upon decreasing levels of land-use intensity was only present for spiders (Table 4 and Fig. 3).

Figure 3.

Alpha diversity of five different arthropod groups in 23 agricultural landscape test sites in relation to agricultural land-use intensity and average proximity index of the semi-natural habitat patches (black circles, Apoidea; white circles, Araneae; black triangles, Carabidae; white triangles, Heteroptera; black squares, Syrphidae). Graphs represent results conditional upon the effect of other variables.

beta diversity

Across taxonomic groups, land-use intensity and percentage of the landscape covered with natural patches, patch proximity and both measures of habitat diversity were significantly related to β diversity negatively and positively (Table 3). Patch proximity explained the variation in β diversity best.

Incorporating the different variables into one model revealed that land-use intensity and patch proximity comprised the set of variables that explained most variation in β diversity. While the effect of patch proximity on β diversity did not differ between taxonomic groups, land-use intensity had a strong homogenizing effect on local bee species composition (Table 4 and Fig. 4).

Figure 4.

Beta diversity of five different arthropod groups in 23 agricultural landscape test sites in relation to agricultural land-use intensity and average proximity index of the semi-natural habitat patches (black circles, Apoidea; white circles, Araneae; black triangles, Carabidae; white triangles, Heteroptera; black squares, Syrphidae). Graphs represent results conditional upon the effect of other variables.

relative contribution of β diversity to γ diversity

On average, between 49% and 60% of the total species diversity of the landscapes could be attributed to β diversity. These proportions differed significantly between the taxonomic groups and were highest for bugs, followed by spiders, carabid beetles, hoverflies and bees (Table 3).

The percentage of the landscape covered with semi-natural habitat patches, patch proximity and both measures of habitat diversity affected β diversity more strongly than α diversity in a positive direction across taxonomic groups. The effect of land-use intensity was only significant for bees (Table 4 and Fig. 5).

Figure 5.

Relative contribution of α diversity to γ diversity for five different arthropod groups in 23 agricultural landscape test sites in relation to habitat diversity of the semi-natural habitat patches (black circles, Apoidea; white circles, Araneae; black triangles, Carabidae; white triangles, Heteroptera; black squares, Syrphidae). Graphs represent results conditional upon the effect of other variables.

The change in relative contribution of β diversity could be explained best by habitat diversity, with all investigated invertebrate groups equally affected and an additive effect of land-use intensity for bees.


This large-scale study demonstrates that the total species richness of arthropods in temperate European agricultural landscapes decreases with increasing management intensity of the agricultural fields and an altered landscape structure. Although both factors are related, in practice their effects appear to be, at least partially, additive. This is emphasized by the fact that the measure of land-use intensity used in the present study is based on characteristics independent of landscape structure (Herzog et al. 2006). The landscape factor that best explains the observed γ diversity is proximity index, which integrates the size and interdistance of the semi-natural habitat patches, but it does not allow separation of the effects of amount of available habitat and habitat isolation. Notwithstanding this fact, as other landscape metrics (patch density, average patch size and total proportion of the landscape covered with patches) did not explain a comparable amount of the unexplained variance, it is likely that both the amount and spatial configuration of suitable habitat are important determinants of biodiversity in the natural habitat patches.

When focusing on biodiversity at a local scale (i.e. α diversity), previously reported negative effects of increased land-use intensity on the local diversity (Duelli, Obrist & Schmatz 1999; Fauvel 1999; Marc, Canard & Ysnel 1999) are confirmed by our data. Negative effects of decreasing proximity upon farming intensity were significant for spiders and reveal that the local spider assemblages of a landscape with small and isolated patches contain fewer species. Local spider assemblages of agricultural fields rarely contain unique species compared with the border of agricultural fields and semi-natural habitat patches. In contrast, more natural habitats embedded in an agricultural matrix contain the majority of the spider species inhabiting the agricultural fields plus a supplement of species bound to the more natural habitat patches (Maelfait & De Keer 1990; Schmidt & Tscharntke 2005). This suggests that local extinction/colonization processes of the latter species group plays an important role in determining local spider species richness. Stochastic effects can be expected to be much higher in small patches and may drive local populations to extinction rather quickly (Hanski 1998). If dispersal capacities of these species are insufficient to cover the distance between these patches, small-scale extinctions are not compensated for by recolonization from nearby patches. For spiders, dispersal occurs mainly by ballooning, in which juveniles produce long silk threads that carry them up in the air currents (Bell et al. 2005). Dispersal propensity differs strongly among spider species and has been shown to correlate positively with the degree of habitat generalism (Bonte et al. 2003; Bell et al. 2005), suggesting that more specialized species become depleted from these isolated patches that consequently contain only a few, generalist species with high dispersal abilities. Thus, local spider species diversity is enhanced in patches surrounded by a larger percentage of non-crop areas, as reported elsewhere (Clough et al. 2005; Schmidt et al. 2005). In contrast, for other arthropod groups, such as carabid beetles, Cole et al. (2005) showed that local species diversity in agricultural fields did not differ significantly compared with natural areas because of a high turnover of typical agrobiont species.

Most importantly, our results suggest that the loss in species richness of the total landscape is not solely the result of a decrease in species richness of the local communities but additionally caused by a loss of species turnover between local communities. Our results imply that local communities of landscapes consisting of small and unconnected patches are characterized by a species composition that hardly diverges between patches. This homogenizing effect can be understood most clearly when combined with the results on α diversity, where we found a clear impoverishment of the local arthropod diversity with increasing land-use intensity and, to a lesser extent, a decreasing proximity index. This suggests that local communities become unsaturated, most probably because of the depletion of specialist, and typically more competitive, species (Tilman et al. 1994) primarily characterized by low dispersal capacities (de Vries, den Boer & van Dijk 1996). These vacant positions within the local community can consequently be replaced by species of the surrounding agricultural matrix. Although these species are typically characterized by low competitive abilities, their high mobility makes them more likely to occupy patches not yet recolonized by superior competitors because of the others’ low dispersal properties (Nee & May 1992; Tilman et al. 1994; Kareiva & Wennergren 1995). This is also suggested by community analysis of the study sites, where it was found that landscapes with small, isolated patches embedded within a highly agricultural intensive matrix contained small and relatively unspecialized species (Aviron et al. 2005; Schweiger et al. 2005).

Effects of increasing land-use intensity upon the effect of proximity on β diversity could only be demonstrated for bees, which decreased from 25 species in less intensively managed landscapes to approximately 10 species in the most intensively managed landscapes. The prominent homogenizing effect of bee species composition with increasing land-use intensity is likely to be linked to their dependency on flowering plant diversity (Tscharntke, Gathmann & Steffan-Dewenter 1998). Our measure of land-use intensity is based on nitrogen input, amount of herbicides and pesticides used and livestock densities. Highly intensively managed landscapes, which consist of a mixture of species-poor grassland, dominated by species such as Lolium perenne and Holcus lanatus, and arable land, are therefore characterized by flowering perennial plants. These management practices also lead to spread of fertilizers and herbicides into the borders of the semi-natural patches, which will subsequently decrease quality in terms of plant species diversity. These environmental circumstances are likely to favour only a small subset of the original bee fauna, such as some generalist bumblebee species that are capable of exploiting mass flowering crops (Westphal, Steffan-Dewenter & Tscharntke 2003).

In relation to habitat diversity, no net effect of increase in species richness was observed in either γ or β diversity. However, habitat diversity caused a shift in the relative contribution of α and β diversity to γ diversity, wherein increasing habitat diversity resulted in an increase of the relative contribution of β diversity. However, it must be emphasized that this will simultaneously lead to a decrease in the contribution of α diversity, probably caused by the decreasing proximity of patches of a particular habitat type.

The coincidence of patterns of β diversity with those of γ diversity stresses the importance of dissimilarity between local community species composition as one of the most important determinants of total landscape species diversity in agricultural landscapes. It is therefore important to emphasize that evaluation and implementation of agri-environment schemes must incorporate the expected gain in species richness as a result of species differences between local communities. When restricting the results at the local scale, for instance, the importance of proximity could only be demonstrated for spiders, while all species groups appeared to be affected by this landscape feature when focusing on β and γ diversity. Evaluation of agri-environment schemes based on the different components of landscape species diversity hence might help to interpret the previously recorded ambiguous results.


We are particularly grateful to the following who helped identify the species: Tim Adriaens, Frank Burger, Rafaël De Cock and Jaan Luig (wild bees), Berend Aukema, Roland Bartels, Jean-Yves Baugnée and Ralph Heckman (bugs), Konjev Desender, Ringo Dietze and Keaty Maes (carabid beetles), Martin Musche and Dieter Doczkal (hover flies) and Herman De Koninck, Valerie Vanloo and Johan Van Keer (spiders).

 Luc Lens, Jan Bengtsson and two anonymous referees are acknowledged for their constructive comments on a previous version of this manuscript. Funding was received from the Energy, Environment and Sustainable Development Programme (FP5) of the European Commission (contract number EVK2-CT-2000-00082).