Grazing intensity affects insect diversity via sward structure and heterogeneity in a long-term experiment

Summary

1. In the past, insect diversity in grasslands showed a severe decline due to management intensification or abandonment. In this study, we investigate the long-term influence of grazing and the potential for spatial patterns created by different grazing intensities to enhance insect diversity.
2. In a long-term experiment (2002–2011), three grazing intensities were applied to 1-ha paddocks in a triplicate block design: moderate grazing (MC), lenient grazing (LC) and very lenient grazing (VLC, since 2005). The experiment was conducted in a moderately species-rich grassland at the edge of the Solling Uplands in Lower Saxony, Germany. Orthoptera (grasshoppers) and Lepidoptera (butterflies) on three 50-m transects per paddock were counted in 2002–2004 and again in 2010 and 2011. Statistics were performed using linear mixed modelling.
3. Grasshopper diversity measures (species richness and abundance) were significantly affected by grazing intensity; abundance increased from 2002 to 2011 more strongly in the LC than in the MC treatment. Butterfly species richness response to grazing intensity varied among years. Data from 2010 and 2011 did not reveal any advantage of the lowest grazing intensity (VLC) compared to the intermediate grazing intensity treatment (LC) in either insect group.
4. Multiple regressions were used to investigate diversity patterns. Along with compressed sward height, spatial patchiness was important for grasshopper species richness and abundance as well as for butterfly species numbers. Butterfly abundance was mainly influenced by vertical sward height heterogeneity in addition to the significant effects of thistle abundance and number of nectar plant species.
5. Synthesis and applications. Cattle grazing intensity affects the proportions and spatial heterogeneity of short and tall sward patches on pastures. The less mobile grasshoppers particularly benefitted from the structural modifications created by cattle at lenient grazing levels (stocking rate 1·14 SLU ha⁻¹, standard livestock unit (SLU) = 500 kg). In the final study years, areas with intermediate grazing intensity revealed high diversity indices and the most distinct patchiness, therefore a further reduction in grazing intensity is not recommended. This indicates that commercial livestock production may be compatible with conservation targets.

Introduction

European grasslands hold a broad diversity of plants and insects and their conservation is considered as a major challenge for land management and nature conservation (Littlewood, Steward & Woodcock 2012). Against the background of decreased biodiversity in the past (Vickery et al. 2001; Stoate et al. 2009), appropriate management of remaining grassland sites is required to maintain biodiversity. Grazing appears to have a high potential for combining these targets with the growing social demands for animal welfare (Van den Pol-van Dasselaar et al. 2008). However, the main function of pastures for farmers is to meet agronomic and financial interests. Therefore, the identification of a threshold grazing intensity that fulfils both environmental and livestock production objectives is essential. It is well known that high stocking rates combined with intensive grassland management contribute to the deterioration of insect diversity (Vickery et al. 2001; Stoate et al. 2009). On the other hand, terminating agricultural activity also leads to a decrease in diversity during succession to woodland (Erhardt 1985; Stoate et al. 2009). However, it is still uncertain what level of grazing intensity is appropriate to conserve insects and by which mechanisms grazing intensity affects insect diversity. Main predictors of explanatory value could include plant diversity (Haddad et al. 2009), sward height (Lawton 1983; Kruess & Tscharntke 2002) and sward structure (Rook et al. 2004; WallisDeVries et al. 2007). As extensive grazing is supposed to enhance phytodiversity (Marriott et al. 2009; Marion, Bonis & Bouzillé 2010), this should in turn promote species richness of higher trophic levels (Siemann 1998). Furthermore, low grazing intensities result in taller swards (Isselstein et al. 2007), providing more forage and shelter for herbivorous insects (Gardiner et al. 2002). When enough forage is available, a heterogeneous sward structure with short and tall patches (defined by Adler, Raff & Lauenroth 2001) emerges (Milchunas, Sala & Lauenroth 1988; Correll, Isselstein & Pavlu 2003; Dumont et al. 2007), offering different ecological niches for many insect species (WallisDeVries et al. 2007).

Butterflies and grasshoppers are often used as biodiversity indicators, for example in the European Grassland Butterfly indicator, which is one of the standard European indicators on biodiversity (Van Sway et al. 2013). Butterflies are well studied (e.g. Ebert 1991), easy to identify and react to changing environmental conditions quickly (Erhardt 1985). As Thomas (2005) stated, they serve as indicators for many insect groups. Grasshoppers have also become one of the most important invertebrate indicators for agricultural management and disturbance (Weiss, Zucchi & Hochkirch 2012). Compared to butterflies, they are less dependent on specific host plants, but are sensitive to changes in vegetation structure (Wettstein & Schmid 1999; Gardiner et al. 2002; Weiss, Zucchi & Hochkirch 2012). Due to their inferior mobility compared to butterflies (Wettstein & Schmid 1999), they depend more on good quality habitats on a smaller scale (e.g. Weyer, Weinberger & Hochkirch 2012).

Previous studies investigating the effect of grazing intensities on insect diversity often only compared two stocking rates (Kruess & Tscharntke 2002; WallisDeVries et al. 2007; Sjödin, Bengtsson & Ekbom 2008), and the spatial heterogeneity of pastures is rarely considered. In addition, data from long-term grazing studies on controlled experiments are very rare and even less frequent in combination with insect recordings.

The aim of this study was to investigate the impact of grazing intensity on the diversity of grasshoppers and butterflies as indicators for faunistic diversity and to elucidate the underlying factors modified by grazing. To this end, a long-term field experiment was carried out with three grazing intensities. The following hypotheses were addressed: I) insect diversity (grasshopper and butterfly species richness and abundance) is driven by grazing intensity. II) This relationship is mediated by a) phytodiversity, b) sward height or c) sward heterogeneity. III) Owing to the formation of stable patches, insect abundance can increase in the long term on the more extensive grazing treatment.

Materials and methods

Experimental site

The study area is situated in Relliehausen (51°46′55″N, 9°42′13″E, 250 m a.s.l.), Lower Saxony, Germany. The soil type is a brown earth/pelosol and the vegetation association a moderately species-rich Lolio Cynosuretum. No fertilizers were used on this site. For more details see Isselstein et al. (2007) and Wrage et al. (2012).

The grazing experiment, previously part of the EU project FORBIOBEN (Isselstein et al. 2007), consisted of three grazing intensity treatments. Grazing intensity modifies average sward height (Isselstein et al. 2007), and so target sward heights were used as a proxy for grazing intensity [moderate stocking (MC): 6 cm, lenient stocking (LC): 12 cm, both set up in 2002 and very lenient stocking (VLC): 18 cm, set up in 2005] and achieved by using a put and take system with adjusted numbers of Simmental cattle. To monitor grazing intensity, compressed sward height (CSH) was measured biweekly (50 measurements per paddock) with a rising plate meter (Castle 1976). The study was conducted in a randomized block design with three replicates each on 1-ha paddocks. For stocking details see Table 1. Weather data are summarized in Wrage et al. (2012).

Table 1. Overview of stocking data (means ± SD) per ha according to grazing intensity treatments [moderate grazing (MC), lenient grazing (LC), very lenient grazing (VLC)] and years used in this investigation. The number (no.) of animals refers to times when paddocks were actually grazed. Standard livestock unit (SLU) = 500 kg

	2002	2003	2004	2010	2011	Mean
Grazing days
MC	134 ± 0	90 ± 0	154 ± 14	75 ± 0	98 ± 2	110 ± 3
LC	134 ± 0	64 ± 0	125 ± 0	70 ± 0	55 ± 0	90 ± 0
VLCa				52 ± 0	55 ± 0	54 ± 0
No. of animals
MC	5·9 ± 0·7	2·8 ± 0·3	4·3 ± 0·8	3·6 ± 0·0	4·1 ± 0·1	4·1 ± 0·4
LC	3·4 ± 0·1	1·1 ± 0·0	2·2 ± 0·0	2·8 ± 0·0	3·0 ± 0·0	2·5 ± 0·0
VLCa				2·0 ± 0·0	2·0 ± 0·0	2·0 ± 0·0
SLU-grazing days
MC	551 ± 71	330 ± 53	523 ± 124	368 ± 17	437 ± 24	442 ± 58
LC	353 ± 85	131 ± 7	292 ± 3	268 ± 7	228 ± 9	255 ± 22
VLCa				148 ± 6	161 ± 2	155 ± 4
Stocking rate (SLU ha^−1)b
MC	2·59 ± 0·33	1·55 ± 0·25	2·48 ± 0·58	1·73 ± 0·08	2·05 ± 0·11	2·08 ± 0·27
LC	1·37 ± 0·06	0·62 ± 0·04	1·40 ± 0·01	1·26 ± 0·03	1·07 ± 0·04	1·14 ± 0·04
VLCa				0·70 ± 0·03	0·76 ± 0·01	0·73 ± 0·02

• ^a Treatment VLC was introduced in 2005. Between 2002 and 2004, paddocks were managed like the LC treatment, but with a traditional cattle breed (German Angus), see Isselstein et al. (2007).
• ^b Stocking rates refer to a stocking season from 1 April to 31 October each year.

Insect recordings

Three permanent 50-m transects per paddock were set up for observation of grasshoppers and butterflies. Transects were monitored in 2002–2004 (see WallisDeVries et al. 2007) and revisited in 2010 and 2011.

Butterfly censuses followed the method of Pollard (1977), in which a 5-m corridor around the monitored transect is imagined (butterfly transect method). Butterflies were recorded biweekly between July and September and identified visually to species level, except for the whites (Pieris brassica, P. napi, P. rapae), which were combined. The whites are widespread generalists that seek crucifers for oviposition and use open flowery areas (Ebert 1991). If individuals were not identified during the census, specimens were net-captured for identification. Butterfly observations were only carried out at appropriate weather conditions (see WallisDeVries et al. 2007). Missing data points due to bad weather were estimated according to the method by Hall (1981).

Grasshoppers were counted using the sweep-net method with one sweep every 2 m (for details see WallisDeVries et al. 2007). Specimen were counted and identified to species level, except for the species hard to distinguish, that is, Chorthippus biguttulus, C. brunneus, und C. mollis, which were pooled. While C. biguttulus is abundant in Germany and can be found on moderately dry habitats, C. brunneus and C. mollis prefer drier areas like dry grasslands (Bellmann 2006) and are therefore not expected to be present frequently in this experiment. Grasshopper recordings were conducted once per month between July and September on dry and more or less sunny days.

Vegetation and sward structure

Vegetation was analysed from 2002 to 2011 on 10 permanent quadrats (1 m² each) per paddock twice per year (spring and autumn). For more information see Wrage et al. (2012). Here, data from 2010 and 2011 are used. Plant species richness and evenness (Magurran 2004) as well as the proportions of functional groups (grasses, forbs and legumes) were calculated for each year. Nectar plant species of the recorded butterfly species as well as host plants were identified using the ‘British butterfly host-plant and nectar source’ data base (http://pbh-butterflies.yolasite.com/hostplants-and-nectar.php, last accessed 01 November 2012). For this measurement, all permanent quadrats per paddock were pooled.

Every 50 cm along the middle transect line, sward surface height was measured with a sward stick (Bircham 1981). The average of these recordings was used as the sward surface height of each paddock. Furthermore, sward height classes (eight classes, in steps of 5 cm each) were counted to gain information about the distribution of microclimates. The evenness (Magurran 2004) of this measure was calculated on the basis of the eight height classes. Openness of the sward was assessed as percentage of bare soil cover in a 15 cm radius around the transect points. Flowering thistles (Cirsium arvense and C. vulgare) were counted continuously along the butterfly transect-corridor around the middle transects due to their importance as nectar resources (e.g. WallisDeVries, Van Swaay & Plate 2012). Despite a positive effect on butterflies, from an agronomic point of view, thistles are undesirable and known to spread into gaps created by intensive cattle grazing (Silvertown & Smith 1989).

Spatial heterogeneity

Proportions of short and tall sward patches were calculated based on aerial photographs (Geobasisdaten der Niedersächsischen Vermessungs- und Katasterverwaltung 2010). The photograph was taken in early spring 2010, when senescent plant material of tall patches and young plant growth of short patches made the spatial heterogeneity clearly visible as colour differences. The image was split into colour channels, and the average grey value (red channel) of the whole experimental area was determined as the threshold distinguishing between patch types. On the basis of this value, proportions of tall patches were calculated for each paddock (Fig. 1).

Figure 1

Image processing for the determination of short and tall patch proportions on an LC paddock using imagej. Top panel: orthophoto (red channel). Bottom panel: the assessment of short (black) and tall (white) patches. In addition, the bimodal sward height distribution reflected in grey values is shown.

As heterogeneous pastures often display a bimodal sward height distribution (Gibb & Ridout 1988; Parsons & Dumont 2003), histograms of grey values for each paddock were evaluated to be uni- or bimodal (Fig. 1). These spatial analyses were performed with imagej (version 1.44p; Abramoff, Magalhaes & Ram 2004).

The coordinates of the transects were connected to the orthophoto in ArcMap (version 10.0, ESRI, Redlands, CA, USA). Patches (short or tall) were then digitalized within buffers around each transect analogous to the size of butterfly transect-corridors. The sum of patches of the three transects per paddock was taken as a measure of patchiness. As sward patches are relatively stable under constant grazing management (Marion, Bonis & Bouzillé 2010; Rossignol et al. 2011; Dumont et al. 2012), we used the values of the 2010 spatial analysis for 2011 as well.

Statistical analysis

In order to account for the additional VLC-treatment set-up in 2005, we focused in the statistical analysis separately on the long-term data set (treatments MC and LC, years 2002–2004, 2010, 2011) or on the three treatments data set (treatments MC, LC, VLC; years 2010 + 2011), respectively. For hypothesis I, butterfly and grasshopper data were analysed in response to grazing intensities concerning both data sets. Species and abundance data of insect censuses were pooled per transect annually, leading to three annual values per paddock. For both response variables, mixed anova models were built for the long-term data set with the factors grazing intensity and year as fixed terms. The random term accounted for the nesting structure of the experimental design (transects nested in paddocks nested in blocks) and for the repeated measures over years. As adding random slopes did not improve models as measured by Akaike Information Criterion (AIC, Akaike 1973), random intercept models were used. In the same way, an autocorrelation structure (corAR1) was tested in each model to account for similarities between adjacent years, but proved to be an adequate adjustment only in the butterfly abundance model. In order to test all three grazing intensities, data of 2010 + 2011 were considered in mixed anova models (structure as above). All analyses were carried out for grasshoppers and butterflies separately.

Concerning years 2010 and 2011, separate anovas were performed to test the effect of treatment on the measured sward variables. These variables with the addition of bimodality of sward height distribution were correlated with species richness and abundance of grasshoppers and butterflies in separate linear models. Host plant availability for butterflies in 2011 was analysed in an anova with treatments as predictor variable.

For hypotheses II, multiple regressions based on mixed models (random structure as above) were performed for eligible sward variables and models simplified by stepAIC procedure. The proportion of tall patches was not integrated due to the high correlation with CSH.

For hypothesis III, long-term data set models were used analogously to the models for hypothesis I, but the fixed term year was taken as a numerical variable. In addition, two species with conservation value were chosen to reflect abundance over the long term: the water-meadow grasshopper Chorthippus montanus is an endangered grasshopper in many European countries (Bellmann 2006), and the marbled white Melanargia galathea is an appropriate butterfly indicator species for traditionally managed grassland (Hampicke 2013). Therefore, abundance data were pooled over transects per paddock. Mixed models accounting for repeated measurements over years were performed with treatments and years (numerical) as fixed terms. The M. galathea model was optimized by an autocorrelation structure (corAR1).

All models were visually checked for meeting model assumptions and transformations as well as variance modelling applied where necessary (adjustments are mentioned in the respective tables).

Analyses were performed in r (R Development Core Team 2012), for mixed models in connection with package nlme (Pinheiro et al. 2012).

Results

In total, 3384 grasshoppers across nine species and 2323 individual butterflies of 20 species were counted. The dominant species within each group were the lesser marsh grasshopper Chorthippus albomarginatus (58%) and meadow brown butterfly Maniola jurtina (47%; see Tables S1 and S2, Supporting Information).

In the long-term data set, significantly more grasshopper species were found on LC than on MC (Fig. 2a, Table 2). Species richness differed among years, but no interaction of the main terms was found. For butterflies, the significant interaction between treatments and years (Fig. 2b, Table 2) suggested that the difference of species richness among treatments varied among years.

Table 2. Results of mixed effects models for butterfly and grasshopper species numbers and abundance in the long-term data set (2002–2011) with grazing intensities moderate grazing (MC) and lenient grazing (LC) and in the three grazing intensities data set including treatments MC, LC and VLC (very lenient grazing) in years 2010 and 2011

	Grasshoppers				Butterflies
	NumDF	DenDF	F	P	NumDF	DenDF	F	P
Long-term data set (2002–2011, two grazing intensities MC, LC)
Species numbers
Treatment	1	2	42·376	0·0228	1a	2	2·303	0·2684
Year	4	64	4·303	0·0038	4	64	63·348	<0·0001
Treatment × Year	4	64	1·534	0·2031	4	64	7·012	0·0001
Abundance
Treatment	1bc	2	76·871	0·0128	1bcd	2	9·244	0·0933
Year	4	64	23·298	<0·0001	4	64	41·361	<0·0001
Treatment × Year	4	64	3·310	0·0158	4	64	1·317	0·2734
Three grazing intensities data set (2010 + 2011, grazing intensities MC, LC, VLC)
Species numbers
Treatment	2e	4	15·081	0·0137	2	4	14·822	0·0141
Year	1	24	10·051	0·0041	1	24	38·531	<0·0001
Treatment × Year	2	24	0·421	0·6609	2	24	0·135	0·8747
Abundance
Treatment	2b	4	16·810	0·0113	2	4	10·006	0·0277
Year	1	24	35·142	<0·0001	1	24	69·778	<0·0001
Treatment × Year	2	24	1·565	0·2296	2	24	1·396	0·2669

• NumDF, numerator degrees of freedom; DenDF, denominator degrees of freedom.
• ^a The variance structure varPower was implemented in the model (using the fitted values in the structure).
• ^b Data were square-root-transformed pre-analysis.
• ^c The variance structure varIdent was used in the model (allowing for differing variances each year).
• ^d The autocorrelation structure corAR1 was implemented in the model (accounting for similar values in adjacent years).
• ^e The variance structure varIdent was used in the model (allowing for differing variances for each treatment).

Figure 2

Mean species richness (species number) per transect of (a) grasshoppers and (b) butterflies as well as abundance (number of individuals) per transect of (c) grasshoppers and (d) butterflies between 2002 and 2011. Error bars show standard errors. Colours of bars refer to treatments: light grey for moderate grazing (MC), medium grey for lenient grazing (LC) and dark grey for very lenient grazing (VLC). Different letters indicate significant treatment effects within years 2010 + 2011 (Tukey's HSD post hoc test). For further statistical results see Table 2.

Abundance analysis of grasshoppers in the long-term data set showed a highly positive effect of LC compared with MC (Fig. 2c, Table 2). Furthermore, the increase in abundance over years was significantly steeper on LC than on MC (years taken as numerical, treatment: P = 0·0128, year: P < 0·0001, interaction treatment × year: P = 0·0059). Butterflies showed a similar, but non-significant trend: there were generally more butterflies counted on LC than on MC (Fig. 2d, Table 2). Slopes of butterfly abundance over time did not significantly differ between treatments MC and LC (years taken as numerical, treatment: P = 0·0913, year: P < 0·0001, interaction treatment × year: P = 0·9911). The long-term consideration of the species C. montanus and M. galathea showed a significantly stronger increase in abundance over years on the LC compared with the MC treatment in both cases (Fig. 3).

Figure 3

Abundance of (a) Chorthippus montanus and (b) Melanargia galathea (transect data per paddock were pooled) on MC (moderate grazing, light grey quadrats), LC (lenient grazing, medium grey triangles) and VLC (very lenient grazing, dark grey dots, only 2010 + 2011). Regression lines refer to MC (light grey line) and LC (dashed medium grey line) and indicate significant interactions between treatment and years (C. montanus: P = 0·0067, M. galathea: P < 0·0001) in mixed models. In both models, years (C. montanus: P = 0·0017, M. galathea: P < 0·0001) and treatments (C. montanus: P = 0·0144, M. galathea: P = 0·0198) were significant. In the model of M. galathea, the integration of a variance structure (varIdent) accounted for different variances per treatment.

Regarding the three treatments in 2010 and 2011, the lowest grazing intensity treatment VLC had more species and more individuals than MC but not more than LC for all grasshoppers and butterflies (Fig. 2 and Table 2). In all diversity analyses, the dependence of diversity measures on year was confirmed (Table 2).

Treatments differed significantly in CSH as well as several other botanical and structural variables (Table 3). Height class evenness did not significantly differ among treatments but tended to be higher on LC and VLC than on MC (P = 0·0573). Likewise, openness of the soil differed by trend among grazing intensities (P = 0·0831) and was smaller on LC and VLC than on MC. Considering the proportion of tall patches and sward surface height, treatment effects were found to be highly significant. Patchiness was significantly larger on LC than on MC, with VLC being intermediate. The cover of legumes was significantly smaller in VLC than in MC, with LC being intermediate. Host plant availability in 2011 did not differ among treatments (P = 0·978).

Table 3. Results of response variables measured in 2010 and 2011 in relation to grazing intensity (mean ± SE) and anova levels of significance for treatment (T, MC, moderate grazing; LC, lenient grazing; VLC, very lenient grazing), year (Y) and their interaction [NS, not significant, (*) P < 0·10, *P < 0·05, **P < 0·01, ***P < 0·001]

Response	MC	LC	VLC	Level of significance
				T	Y	T × Y
CSH (cm)	6·17 ± 0·29^c	9·59 ± 0·32^b	11·70 ± 0·28^a	***	NS	NS
Tall patches (%)b	14·83 ± 6·24^b	61·15 ± 3·10^a	75·35 ± 1·22^a	***
SSHa (cm)	6·62 ± 1·22^b	14·11 ± 1·18^a	20·66 ± 2·38^a	***	NS	NS
PlantSpR (m^−2)	10·58 ± 0·50	10·30 ± 0·49	9·93 ± 0·60	NS	NS	NS
PlantE (m^−2)	73·19 ± 0·90	72·83 ± 1·47	71·87 ± 2·06	NS	NS	NS
Grasses (cover %)	64·33 ± 3·37	65·13 ± 5·17	64·16 ± 4·45	NS	NS	NS
Forbs (cover %)	16·88 ± 3·35	18·88 ± 3·77	17·00 ± 3·38	NS	NS	NS
Legumes (cover %)	7·93 ± 1·18^a	5·70 ± 1·11^ab	3·09 ± 0·55^b	**	NS	NS
Openness (%)a	6·17 ± 0·87	4·08 ± 0·42	3·92 ± 0·57	(*)	NS	NS
NectarN (10 m^−2)	13·17 ± 1·09	14·17 ± 1·72	16·08 ± 0·76	NS	NS	NS
Thistles (750 m^−2)	11·00 ± 5·88	16·17 ± 4·30	4·83 ± 2·39	NS	NS	NS
HN (50 m^−1)	5·17 ± 0·95	6·67 ± 0·49	6·83 ± 0·48	NS	NS	NS
HE (50 m^−1)	0·50 ± 0·10	0·73 ± 0·04	0·76 ± 0·05	(*)	NS	NS
Patchinessb (750 m^−2)	6·00 ± 0·58^b	13·33 ± 1·20^a	8·33 ± 1·76^ab	*

• Different letters show significant differences between treatments (post hoc Tukey's HSD test). CSH, compressed sward height; SSH, sward surface height; PlantSpR, plant species richness; PlantE, plant evenness; NectarN, number of nectar plant species; Thistles, number of thistles; HN, number of surface height classes; HE, evenness of surface height classes.
• ^a The response variable was log-transformed prior to analysis.
• ^b Analysis refers to the aerial image of 2010 and was only tested for treatment effects.

Grey value histograms of the aerial image showed a bimodal distribution in all LC paddocks and two of three VLC paddocks (an example is shown in Fig. 1). The remaining paddocks were unimodal. In separate models of structural variables, the proportion of tall patches and bimodality correlated significantly with all insect diversity measures (Table 4). Botanical species richness showed a significant positive correlation only to grasshopper abundance. Height class evenness was positively associated with number of insect individuals and with grasshopper species richness by trend. Butterfly species richness was related to patchiness, as were grasshopper measures by trend.

Table 4. Results of separate mixed effects models for butterfly and grasshopper species numbers and abundance in relation to sward structure variables. Shown are the estimates (Est.) of remaining variables and corresponding P-values

Variables	Grasshoppers				Butterflies
	Species richness		Abundancea		Species richness		Abundance
	Est.	P	Est.	P	Est.	P	Est.	P
CSH	0·494b	<0·0001	0·759c	<0·0001	0·197	0·2480	1·497	0·0240
PlantSpR	0·262b	0·2302	0·777c	0·0285	−0·561	0·1628	−2·091	0·2962
Openness	−0·106b	0·0671	−0·231c	0·0246	−0·136	0·4347	−0·346	0·7161
NectarN	0·121b	0·2950	0·280c	0·0921	0·181	0·2013	1·193	0·0513
Thistles	−0·019d	0·0612	−0·044c	0·0090	0·075	0·0107	0·430d	0·0003
Tall patches	0·042c	0·0009	0·061e	0·0007	0·038	0·0329	0·168	0·0273
HE	3·218	0·0651	6·435e	0·0029	3·697	0·1205	22·622	0·0279
Patchiness	0·224d	0·0767	0·335e	0·0684	0·269	0·0481	0·932	0·1569
Bimodality	2·308e	0·0052	3·009e	0·0113	2·425	0·0043	10·683	0·0126

• CSH, compressed sward height; PlantSpR, plant species richness; NectarN, number of nectar plant species; Thistles, number of thistles; HE, evenness of surface height classes.
• ^a Data were square-root-transformed pre-analysis.
• ^b The variance structure varPower was implemented in the model (using the fitted values in the structure).
• ^c The variance structure varExp was implemented in the model (using the fitted values in the structure).
• ^d The variance structure varIdent was used in the model (allowing for differing variances for each treatment).
• ^e The variance structure varIdent was used in the model (allowing for differing variances for each year).

Variables that were included in multiple regression models are listed in Table 5. As bimodality correlated strongly with patchiness and tall patches with CSH, these terms were left out of the analysis. The multiple regression of grasshopper species richness revealed the importance of CSH in addition to a significant effect of patchiness. Grasshopper abundance was also mainly affected by CSH, although height class evenness, sward openness and patchiness were also significant. The abundance of thistles and CSH were the best predictors for butterfly species richness. Furthermore, patchiness affected species numbers significantly. In addition to an effect of nectar species numbers, the incidence of thistles and height class evenness had a clear impact on butterfly abundance.

Table 5. Results of simplified multiple regression models for butterfly and grasshopper species numbers (F-statistic) (NumDF, numerator degrees of freedom; DenDF, denominator degrees of freedom)

	Species richness				Abundance
	NumDF	DenDF	F	P	NumDF	DenDF	F	P
Grasshoppers
CSH	1a	24	90·740	<0·0001	1bc	24	118·889	<0·0001
HE	1	24	0·440	0·5132	1	24	4·550	0·0433
Patchiness	1	5	7·039	0·0453	1	5	4·346	0·0479
Openness	–d	–	–	–	1	24	6·817	0·0476
PlantSpR	1	24	2·828	0·1056	–	–	–	–
Butterflies
CSH	1	25	9·909	0·0042	–e	–	–	–
HE	–	–	–	–	1	24	18·427	0·0003
Patchiness	1	5	13·319	0·0148	–	–	–	–
NectarN	–	–	–	–	1	24	7·500	0·0114
Thistles	1	25	10·993	0·0028	1	24	39·613	<0·0001

• CSH, compressed sward height; HE, evenness of surface height classes; PlantSpR, plant species richness; NectarN, number of nectar plant species; Thistles, number of thistles.
• ^a The variance structure varPower was implemented in the model (using the fitted values in the structure).
• ^b Grasshopper abundance was square-root-transformed pre-analysis.
• ^c The variance structure varExp was implemented in the model (using the fitted values in the structure).
• ^d ‘–’ terms were eliminated during model simplification.
• ^e The variance structure varIdent was used in the model (allowing for differing variances for each treatment).

Discussion

The present study clearly confirms the detrimental effects of intensive grazing on butterfly and grasshopper diversity (Kruess & Tscharntke 2002; WallisDeVries et al. 2007; Eschen et al. 2012). Interestingly, the species richness and abundance of insects did not significantly differ and showed even higher values on LC than on VLC in most cases. In the following subsections, we are going to consider the differing possibilities proposed to influence grasshopper and butterfly species richness and abundance.

Phytodiversity

Plant diversity was investigated as a potential predictor as it is presumed to have a bottom-up effect on insect diversity, that is, the more plant species, the higher the insect diversity (Siemann 1998; Haddad et al. 2009). However, our data did not show a clear effect of grazing intensity on plant species richness (Table 3), an observation also found by other studies (Kruess & Tscharntke 2002; a; Eschen et al. 2012). Likewise, botanical species richness did not consistently affect insect indices (Tables 4 and 5). One exception was butterfly abundance, which was positively influenced by nectar resources due to the butterflies' dependence on nectar (e.g. Clausen, Holbeck & Reddersen 2001; WallisDeVries, Van Swaay & Plate 2012). Moreover, Zhu et al. (2012) stated that the positive relationship between plant and insect diversity can be altered fundamentally by grazing.

Although in this investigation only adult insects were recorded, we analysed host plant availability for caterpillars in 2011 but found no difference between treatments.

Sward height

Resource-productivity (Siemann 1998) and resource-diversity hypotheses (Lawton 1983) both suggest that taller swards enhance herbivore diversity by providing more biomass (resource-productivity hypothesis) or more microclimate and feeding niches (resource-diversity hypothesis). In the latter hypothesis, the distribution of sward height classes in this study indicated that the tallest sward did not provide significantly more niches than the intermediate treatment. Thus, although taller swards can potentially provide more diverse microclimates (Dennis, Young & Gordon 1998), the proportion of tall patches at an intermediate grazing intensity of c. 10 cm CSH seems to be adequate to cover the whole array of required vertical sward heterogeneity.

The intermediate-disturbance hypothesis, which predicts the largest diversity at a medium level of disturbance (Connell 1978), seems to provide a better explanation of our data. In line with this hypothesis, Pöyry et al. (2006) showed that butterfly diversity peaked at an intermediate sward height, which was taken as an indicator of grazing intensity. Likewise, the grasshopper species Chorthippus albomarginatus, C. parallelus and C. brunneus were found to prefer an intermediate CSH of 10–20 cm (Gardiner et al. 2002), which is in line with the highest diversity values we found on LC and VLC with CSH values of 10 and 12 cm, respectively (Table 3).

Spatial heterogeneity

The spatial heterogeneity of the sward was found to be a key feature for insect diversity, which is consistent with many other studies of cattle grazing, particularly under continuous grazing (Gibb et al. 1997; Cid & Brizuela 1998; Dumont et al. 2007). Nevertheless, in most studies with reference to insect diversity, sward heterogeneity was determined using measures such as variance, standard deviation, standard error or coefficient of variation (as seen e.g. in Morris, Derry & Hardy 1999; Correll, Isselstein & Pavlu 2003; Cole et al. 2010; Eschen et al. 2012). Even though these measures can describe vertical heterogeneity, they cannot be used sufficiently for the characterization of the spatial arrangement of sward heights in terms of patches. Indeed, these terms are not necessarily connected; for example, Guo et al. (2004) reported that spatial and vertical heterogeneity can be influenced differently by grazing: whereas ungrazed plots showed a higher vertical heterogeneity, grazed plots were spatially more heterogeneous. Instead of using a relative measure like the coefficient of variation, we regarded sward height class evenness as an adequate measure for vertical sward heterogeneity in terms of an even distribution of different microclimates. In addition, we assessed spatial heterogeneity via ArcGIS. In our study, this spatial patchiness was clearly enhanced by the more extensive treatments, in particular by the intermediate grazing intensity. This is not only reflected by the number of patches recorded but also by the bimodal sward height distribution seen in the aerial photograph.

As insect species have different habitat requirements (Guido & Gianelle 2001; Potts et al. 2009), heterogeneous swards can offer adequate conditions for a large array of species (habitat heterogeneity hypothesis, Dennis, Young & Gordon 1998), both applying for butterflies (Clausen, Holbeck & Reddersen 2001) and grasshoppers (Gardiner & Hill 2004). Tall areas experience smaller fluctuations in temperature, but offer more forage and shelter (Gardiner et al. 2002). Managing pastures extensively resulting in a patchy structure would therefore not only favour insect herbivores of short or tall grassland but also species benefitting from both structural forms (Gardiner et al. 2002; WallisDeVries et al. 2007; Cole et al. 2010; Weiss, Zucchi & Hochkirch 2012). For these reasons, the creation of a heterogeneous sward is regarded as one of the key issues for preserving biodiversity of grasslands (Clausen, Holbeck & Reddersen 2001; Benton 2003; Rook & Tallowin 2003; Dumont et al. 2009), which can be clearly confirmed in this study.

Long-term development

Our results showed an enhanced long-term development of grasshopper abundance on LC plots. This could be due to the higher patch stability over years on LC in contrast to MC. On intensive pastures, tall patches occur exclusively around dung pats and are, as a result, ephemeral (Gibb & Ridout 1988). Though these areas can act as refugia for arthropods, they are isolated islets (Helden et al. 2010) not habitat networks. On the contrary, tall patches on extensive pastures are mainly maintained due to lower forage quality (Adler, Raff & Lauenroth 2001; Dumont et al. 2007) and are thus more stable and persistent across years (Marion, Bonis & Bouzillé 2010; Rossignol et al. 2011; Dumont et al. 2012). Consequently, there is a larger probability that oviposition sites are not destroyed and offer forage and shelter for the next generations. Furthermore, individuals living on these more or less stable patches do not suffer from habitat fragmentation and changing conditions mediated by intensive grazing or altered management. This would be detrimental in particular for the endangered grasshopper C. montanus, the second most-dominant species in our experiment, which is strongly dependent on its local habitat and not able to cross large distances (Weyer, Weinberger & Hochkirch 2012). Our data strongly support the advantage of continued long-term management on this species and also for the butterfly M. galathea (Fig. 3).

Additionally, stable patches offer different conditions for plant survival and establishment, thereby enabling the development of distinguishable patch-dependent vegetation compositions (Correll, Isselstein & Pavlu 2003; Marion, Bonis & Bouzillé 2010; Wrage et al. 2012). This could further enhance diverse insect assemblages and would thus constitute an added effect of patchiness on biodiversity on top of the more structural differences.

Management suggestions

Pasture management should aim for a heterogeneous sward structure with a slight emphasis on tall patches, which can be achieved at intermediate grazing intensity (stocking rate: ~ 1 SLU ha⁻¹ under the conditions of this study). This management regime provides diverse microhabitats for grasshoppers and butterflies, which, in turn, should also enhance higher trophic levels of biodiversity (Vickery et al. 2001). In order to meet these aims, further reduction in stocking rates is not required, as our results have shown and therefore a sustainable combination of conservation issues and agronomic targets can be achieved. However, larger pastures and animal herd sizes can produce results that differ from those of this study. Here, we used relatively small paddocks with only a few grazing individuals, and social interactions across fences were likely. Nonetheless, paddocks of 1-ha in size are not unusual in the study region, where grassland plots are smaller than 2 ha on average (e.g. Klimek et al. 2008).

Continuous extensive grazing systems should be managed in the same way across years. This management regime can be very valuable for the long-term insect population development by providing secure habitat networks for less mobile, structure-sensitive insects such as grasshoppers.

Acknowledgements

The research leading to these results has received funding from the European Community's Seventh Framework Programme (FP7/2007-2013) under the Grant Agreement No. FP7-244983 (MULTISWARD). The field experiment was initiated and first results from 2002 to 2004 obtained within the EU project QLK5-2001-00130 ‘Integrating foraging attributes of domestic livestock breeds into sustainable systems for grassland biodiversity and wider countryside benefits (FORBIOBEN). We gratefully acknowledge our technician Barbara Hohlmann and our partners at the experimental farm Relliehausen, especially Arne Oppermann and Knut Salzmann. We are very grateful to Michael Sayer (deceased) for his great effort in insect species identification in the earlier years of this study. Furthermore, we thank Christoph Scherber for statistical advice.

Data accessibility

Data available through DRYAD entry doi: 10.5061/dryad.d9q57 (Jerrentrup et al. 2014).