Enhancing invertebrate food resources for skylarks in cereal ecosystems: how useful are in-crop agri-environment scheme management options?

Authors


*Corresponding author. E-mail: bsmith@gct.org.uk

Summary

  • 1UK agri-environment schemes rarely address within-crop biodiversity yet this habitat is used almost exclusively by some taxa. With the removal of set-aside, maximizing the ecological services provided by in-crop management options is critical. Field trials of small undrilled patches in the cropped area have been shown to increase skylark Alauda arvensis numbers and breeding productivity. This response may reflect a benefit for lower trophic levels, such as the invertebrates and arable plants on which birds feed. The technique has now been adopted as an option in the Environmental Stewardship scheme in England.
  • 2Two within-field management techniques (undrilled patches and wide-spaced rows) were compared with standard row spacing on 10 conventionally managed farms growing winter wheat in northern and eastern England in 2002 and 2003. The effect of the treatments on invertebrate abundance, particularly invertebrates known to be important food for birds, was compared and the link between cover of arable weeds and invertebrate abundance was assessed. The diet of chicks located on the treatments was compared by faecal analysis.
  • 3On a field scale, the treatments did not consistently increase invertebrate numbers. Invertebrates either colonized or avoided undrilled patches in the crop. The extent to which patches were colonized was dependent on vegetation cover, with the invertebrate assemblage structure responding strongly to broadleaved weed cover and, to a lesser extent, grass cover.
  • 4The diet of skylark chicks did not differ between nests located in the two treatments.
  • 5 Synthesis and applications. Neither undrilled patches nor wide-spaced rows benefited invertebrate populations at the field scale. The value of undrilled patches for invertebrates would be enhanced by promoting weed cover, particularly broadleaved weeds; this could be achieved by creating the patches in the spring using cultivation and avoiding spraying with broad-spectrum herbicides. Higher numbers of non-pernicious weeds could provide food and refuge for invertebrates which are a food source for skylarks and other farmland birds that forage in the crop.

Introduction

In the second half of the 20th century, intensification of farming methods led to nationwide agricultural change (Potts 1997) which in turn contributed to a decline in invertebrate diversity on arable land (Aebischer 1991). For invertebrates inhabiting arable crops, there is evidence of a decline for polyphagous predators, Staphylinidae (rove beetles), Chrysomelidae (leaf beetles), Hymenoptera (parasitoid wasps), Lepidoptera (moths, localized butterfly species), Araneae (spiders) and Opiliones (harvestmen) (Aebischer 1991; Ewald & Aebischer 1999; Conrad et al. 2004). This decline has been attributed both to increased use of insecticides and to more efficient weed control (Ewald & Aebischer 1999; Wilson et al. 1999). Incorporating weedy patches within the crop may increase invertebrate diversity and abundance, as has been found when cereal field margins are selectively sprayed (Chiverton & Sotherton 1991; Moreby & Southway 1999). Weeds provide a direct resource for phytophagous species, serve as an indirect resource for predatory species and add structural diversity (Speight & Lawton 1976; Hawes et al. 2003). The decline of some farmland birds has been causally linked to limited invertebrate food supplies (Potts & Aebischer 1995; Brickle et al. 2000; Hart et al. 2006). It has been demonstrated that loss of food (especially invertebrate chick-food) and nesting habitat in the cropped areas of the agricultural landscape are key drivers in farmland bird declines (Butler, Vickery & Norris 2007).

Uptake in the English agri-environment Environmental Stewardship (ES) scheme shows a bias towards boundary feature and margin management options (Rural Development Service 2006). However, with an increase in the production of bio-fuels and a reduction in land that is ‘set-aside’ from production, the amount of farmland under non-crop habitat may be reduced, and consequently, in-crop management options will be necessary to increase resources for farmland wildlife (Storkey & Westbury 2007). Whether conservation strategies designed for birds tangentially benefit invertebrates is rarely tested, but given the decline of arable plants and invertebrates on farmland, it is important that the options in agri-environment schemes benefit multiple desired species (Olson & Wackers 2007). The experiment we describe here has already demonstrated that leaving undrilled patches in wheat fields can lead to an increase in the breeding success of skylarks Alauda arvensis by 0·5 chicks per breeding attempt (Morris et al. 2004). In this study, we compare the effect that undrilled patches and wide-spaced rows have on invertebrate populations and we investigate the links between vegetation cover, invertebrate abundance and skylark diet.

Materials and methods

Field trials were carried out April–August 2002 and 2003, on 10 sites sown with winter wheat, located in northern and eastern England. There were three treatments on each site: (i) conventional husbandry (CONV), the experimental control, with normal row spacing of 12·5 cm between rows; (ii) undrilled patchwork (UDPW); patches were c. 4 × 4 m, at a density of two patches per hectare; (iii) wide-spaced rows (WSR), double the normal row width (but with the same rate of seed sown as CONV).

Fields were selected with characteristics likely to maximize densities of skylark (Wilson et al. 1997); each field was > 5 ha and had a relatively open aspect, with minimal influence from surrounding tall hedges, tree lines and woodland.

Insects were collected by vacuum sampling and pitfall trapping along with plant monitoring at the sampling locations. Vacuum sampling was carried out using a Dvac suction sampler (Model 1-A, D-Vac Company, Ventura, CA, USA) (Dietrick 1961) in May, June and July of each year; each sample comprised five 10-s sessions and sampled an area of 0·092 m2. Pitfall trapping was conducted using a 6-cm diameter, white plastic pitfall trap half-filled with 50% ethylene glycol (antifreeze) and unscented detergent. Two pitfall traps were installed 2 m apart at each sampling location and were opened for 7 days in mid-June. The contents were pooled for analysis. In the CONV and WSR treatments, samples were taken at eight randomly selected locations within the crop (excluding a 30-m headland), whereas in the UDPW treatment, eight randomly selected undrilled patches (UP) were sampled along with eight randomly selected locations in the crop surrounding the undrilled patches (CropUP). Invertebrates were identified to family. Plant species composition was assessed on two occasions in mid-May and early July. Twenty-four quadrats, each 0·25 m2, were sampled from each treatment, plus an additional 24 UP quadrats in the UDPW treatment. Quadrats were placed in eight groups of three using the same location points as the invertebrate sampling. Percentage cover of each plant species was recorded plus crop, bare ground (viewed from below the canopy) and plant litter. Cover was recorded in the following categories, with the midpoint value used for analysis: 0–1%, > 1–2%, > 2–5%, > 5–10%, > 10–20% and then in 10% bands up to > 90–100%.

Faecal samples were collected between April and September 2002 and 2003 from skylark nestlings aged between 6 and 9 days, and from fledglings and adults either when birds were handled or following observations of birds defecating. All samples were stored in 70% alcohol until processed. Identification was to taxonomic order according to Moreby (1988). It was possible to identify some Coleoptera to family, and Carabidae were further classified into categories of small (0–5 mm), medium (5–10 mm) or large (> 10 mm).

The sampled fields varied between years; consequently, years were analysed separately. Invertebrate species data were bulked to give a number of variables (Table 1). At each site, the arithmetic mean of data from the eight samples collected within each treatment was calculated and data were log-transformed prior to analysis. To determine the field-scale effect of introducing UP, a weighted mean was calculated using samples collected from within the UP and the CropUP (the overall treatment is referred to as undrilled patchwork (UDPW)). UDPW reflected the effect of treatment over the whole field and was then directly comparable with samples collected from the WSR and CONV.

Table 1.  Bulked response variables used in univariate statistical analyses to assess the effect of experimental treatments
Source of variationGroup members
Total invertebratesSum total of all invertebrates
CFIGrey partridge chick-food index (Potts & Aebischer, 1991; updated N. J. Aebischer personal communication)
SFISum of skylark food items, derived from faecal analysis
Generalist predatorsGroups within which the species were predominately predatory
PhytophagesGroups within which the species were predominately herbivorous or expected to be pollinators or nectar feeders
HomopteraSum of Homoptera (hoppers)
HeteropteraSum of Heteroptera (true bugs)
CarabidaeSum of carabid (ground beetle) species
Carabid species richnessNumber of carabid species
StaphylinidaeSum of staphylinid (rove beetle) species
Staphylinid species richnessNumber of staphylinid species
Total ColeopteraSum of Coleoptera (beetles)
LycosidaeSum of Lycosidae (wolf spiders)
DipteraSum of Diptera (flies)
Species richnessNumber of species

General Analysis of Variance (anova) was employed to detect treatments effects (CONV, UDPW and WSR). The model was a standard randomized blocks anova; site was specified as ‘block’ and a repeated-measures structure was used in the case of data collected on more than one occasion. To determine the extent to which UP affected the distribution of invertebrates within fields, a separate analysis was run to compare samples collected from within individual patches (UP) with samples collected from the surrounding crop (CropUP). As before, data were analysed using anova with site specified as ‘block’, employing repeated-measures where appropriate. Analyses were carried out using Genstat 8·2 (Genstat 11 Committee 2008). To determine the response of the invertebrate groups to components of vegetation cover (grass cover, broadleaf cover, crop cover, bare ground and litter), we constructed Generalized Linear Models, with Poisson distribution (for count data) and log-link function. The models were refitted using quasi-Poisson errors because overdispersion was indicated in all cases. Response data were calculated using invertebrate abundance per sample collected in June (pitfall) and May/July (Dvac) from both the individual undrilled patches (UP) and surrounding crop (CropUP) in UDPW fields as well as the wide-row (WSR) and conventional (CONV) fields. Separate models were constructed for the following invertebrate groups: (i) pitfall data – total invertebrate abundance, generalist predators, phytophages, Carabidae and Staphylinidae; (ii) Dvac data – total invertebrate abundance, generalist predators, phytophages, CFI (grey partridge chick food index) and SFI (sum of skylark food items). Predictor variables were calculated from plant cover data collected from the same locations as invertebrate data. Pitfall data were collected in June, when no plant assessments were undertaken; therefore, July vegetation cover was used as a surrogate. Structural variables (sampling date for Dvac samples, site/field and treatment) were included in the model. Analyses were carried out using r version 2·7 (The r Foundation for Statistical Computing, 2008).

The invertebrate assemblage was analysed using redundancy analysis. A priori detrended correspondence analysis was carried out on the species data; the gradient lengths were < 3, indicating that linear analysis was appropriate (Ter Braak & Smilauer 1998). Redundancy analysis was used to establish the linear relationships between environmental variables and invertebrate community composition excluding the variation accounted for by site differences. Thirty invertebrate taxa were included in the analyses: Linyphiidae, Lycosidae, Opiliones, Gastropoda, Orthoptera, Aphididae, Homoptera, Heteroptera, Neuroptera, Lepidoptera, Symphyta (adults and larvae separately), Formicidae, Carabidae, Staphylinidae, Chrysomelidae, Curculionidae, Cantharidae, Elateridae, Coccinellidae (adults and larvae separately), other Coleoptera, Tipulidae, Nematocera, Acalypterae, Aschiza, Brachycera, Calyptera and Diptera larvae. Environmental variables were: sampling time (May, June), treatment (CONV, UP, CropUP, WSR) and vegetation components (grass cover, broadleaf cover, crop cover, bare ground and litter cover). The data were untransformed. The significance of the relationship between environmental variables and ordination axes was carried out using 499 Monte Carlo permutations, restricted within site. All environmental variables were included and tested with Monte Carlo permutations before adding by manual selection; structural variables were added first. Data for 2002 (n = 631) and 2003 (n = 535) were analysed separately. Analysis was carried out in Canoco 4·5 (2002).

The diet composition of skylark nestlings located on each of the treatments was compared using compositional analysis (Aitchinson 1986). Diet was categorized into eight invertebrate and one mixed group: Arachnida, Hemiptera, adult Hymenoptera, Lepidoptera, Carabidae, other Coleoptera, Diptera and other food items. The latter category included Orthoptera, snails and cereal grains. Faecal samples were collected from nests located in each of the treatments; data were bulked across nests and field. For a comparison of the proportion of invertebrate groups between years, treatments and months, the proportions were analysed using log ratio analysis (e.g. Cummins & O’Halloran; 2002, Brown & Aebischer, 2003). Nest occurrence varied between months; for example, by July 2002 nests only remained in UDPW. Year, month and treatment were included in the model but no interactions were tested due to this imbalance.

Results

Of the 56 weed species identified and recorded, only 19 were common, and the overall percentage weed cover was generally low. Only Alopecurus myosuroides Huds. (black-grass) was recorded at more than 1% cover during July in either year when averaged across all sites and treatments. However, as would be expected, there was considerable variability between sites and some species were locally common. Differences between years were particularly marked in the undrilled patches where weed cover was much higher in 2002 than in 2003. Comparisons of vegetation cover between UP and CropUP indicated significantly higher weed cover within UP than in CropUP. However, there was also an interaction between treatment and time of sampling (F = 18·81,18, P < 0·001). As the growing season progressed, weed cover increased in UP at a greater rate than in CropUP. Crop cover was lower in UP than CropUP in both years; weed species richness was higher in UP.

field scale effects

In 2002, there were differences in invertebrate species richness, abundance of Staphylinidae (rove) beetles and of Lycosidae (wolf spiders) captured by pitfall trap (Table 2). Abundance of staphylinid beetles was highest in fields with undrilled patches (UDPW), double that which occurred in wide-spaced row fields (WSR) (Fig. 1a). Lycosids were more abundant in WSR and least abundant in conventional fields, but in all treatments, average abundance per sample was < 1 (Fig. 1a). A mean of 9·8 ± 0·8 invertebrate species per pitfall trap were caught in UDPW fields compared with 6·4 ± 0·5 in conventional fields and 6 ± 0·8 in WSR.

Table 2. anova table for analyses of ground-active invertebrate data collected by pitfall trap in 2002 and 2003. The table shows F statistics for treatment effects of (i) conventional (CONV)/undrilled patch (UDPW)/wide-spaced row (WSR) treatments, (ii) individual undrilled patches (UP)/crop surrounding patches (CropUP)
Source of variationPitfall 2002Pitfall 2003
CONV/UDPW/WSR
d.f. = 2, 26; n = 27
UP/CropUP
d.f. = 1, 17; n = 18
CONV/UDPW/WSR
d.f. = 2, 25; n = 26
UP/CropUP
d.f. = 1·17; n = 18
  • *

    P < 0·05;

  • **

    P < 0·01;

  • ***

    P < 0·001; nsd, no significant difference.

Araneae
 Lycosidae3·73*6·13*0·7nsd5·10*
Coleoptera
 Carabidae1·98nsd0·58nsd1·76nsd180·01***
 No. of carabid species1·67nsd2·62nsd2·15nsd167·31***
 Staphylinidae3·67*8·93*0·29nsd29·99***
 No. of staphylinid species2·99nsd20·63**0·01ns0·14nsd
Composite variables
 Generalist predators2·67nsd0·04nsd1·73nsd194·00***
 Phytophages1·48nsd7·91*2·09nsd19·14***
 Total invertebrates2·36nsd0·53nsd1·48nsd209·91***
 Species richness24·42***1·55nsd1·8nsd156·06***
Figure 1.

Mean number of invertebrates in Conventional (CONV), undrilled patchwork (UDPW) representing the weighted mean, and wide-spaced row (WSR) fields, sampled in 2002 by (a) pitfall trap and (b) Dvac, with 95% confidence limits.

In 2002, only ‘generalist predators’ (which included the predatory flies) captured by Dvac differed between the treatments (Table 3), but these were most abundant in conventional fields (Fig. 1b). In general, invertebrate abundance fluctuated over time, significantly so for many species, but there were no significant time × treatment interactions; i.e. data reflected the natural phenology of sampled invertebrates unaffected by treatment. No significant differences were found in 2003 for either pitfall or Dvac-sampled invertebrates possibly because weed cover was 50% lower than in 2002.

Table 3. anova table for analyses of crop-active invertebrate data collected by Dvac in 2002 and 2003. The table shows F statistics for treatment effects of conventional (CONV)/undrilled patchwork (UDPW)/wide-spaced row (WSR) and the differences in these effects over the sampling period (time)
Source of variation20022003
Treatment
(CONV/ UDPW/WSR)
(d.f. = 2, 18)
Time
(May/June/July)
(d.f. = 2, 51)
Treat × time
(d.f. = 4, 51)
Treatment
(CONV/ UDPW/WSR)
(d.f. = 2, 15)
Time
(May/June/July)
(d.f. = 2, 46)
Treat × time
(d.f. = 4,46)
  • *

    P < 0·05;

  • **

    P < 0·01;

  • ***

    P < 0·001; nsd , no significant difference.

Araneae
 Linyphiidae2·33nsd28·01***0·63nsd0·73nsd4·81*0·21nsd
 Coleoptera      
 Carabidae0·96nsd2·70nsd0·21nsd2·00nsd14·11***0·84nsd
 Staphylinidae0·78nsd9·07***0·44nsd0·44nsd18·31***0·33nsd
 Total Coleoptera0·42nsd3·21*0·80nsd2·73nsd4·21*0·88nsd
Diptera0·21nsd13·28***0·50nsd1·32nsd11·55***0·30nsd
Hemiptera
 Heteroptera1·34nsd0·01nsd0·55nsd0·32nsd5·13**0·21nsd
 Homoptera1·13nsd9·12***0·20nsd0·36nsd3·47*0·98nsd
Composite variables
 Generalist predators4·70*30·31***0·20nsd0·61nsd19·77***0·71nsd
 Phytophages0·67nsd23·87***0·09nsd0·24nsd80·1***0·85nsd
 Total invertebrates0·73nsd32·70***0·21nsd1·05nsd12·20***0·34nsd
 Skylark food items (SFI)0·12nsd18·47***0·39nsd1·04nsd11·95***0·35nsd
 Chick food index (CFI)0·96nsd1·94nsd0·49nsd0·85nsd2·41nsd1·05nsd

localized effects of undrilled patches

When samples of ground-active invertebrates from within the undrilled patches (UP) were compared with those from the crop surrounding the patches (CropUP), significant differences in invertebrate abundance were detected in both years (Table 2). In 2002, Staphylinidae abundance and species richness were both higher in the CropUP than within UP, while Lycosidae and phytophagous species were more abundant in UP (Fig. 2a). In 2003, the lack of invertebrates sampled within UP was striking; all groups of ground-active invertebrates were significantly less abundant in UP when compared with CropUP. For example, there was an average of 104 ± 0·4 invertebrates in each pitfall sample taken from CropUP and 0·3 ± 0·07 invertebrates in those taken from UP; this pattern was consistent across groups: generalist predators – CropUP = 97·9 ± 0·4, UP = 0·3 ± 0·07; phytophages – CropUP = 2·3 ± 0·3, UP = 0·02 ± 0·005; Carabidae – CropUP = 95·8 ± 0·5, UP = 0·3 ± 0·07; Staphylinidae – CropUP = 3·3 ± 0·3, UP 0·01 ± 0·004 and species richness – CropUP = 8·0 ± 0·06, UP = 0·03 ± 0·003.

Figure 2.

Mean number of invertebrates in undrilled patches (UP) and the crop surrounding the patches (CropUp), sampled in 2002 by (a) pitfall trap and (b–c) Dvac, with 95% confidence limits.

In 2002, the number of phytophagous invertebrates sampled by Dvac varied between CropUP and UP (Table 4). There were time × treatment interactions for total invertebrates, Heteroptera and CFI. Fig. 2b,c demonstrate the general trend; in May and June, a higher abundance of invertebrates resided in CropUP rather than in UP, but in July as the crop ripened and weed cover in UP became better established, invertebrates colonized the weedy patches. Heteroptera occurred in very low numbers (< 1 0·5 m−2). In contrast, numbers of invertebrates within the patches remained negligible throughout the season in 2003. In July, the average total number of invertebrates per 0·5 m2 was 25·8 ± 0·3 in CropUP and 0·4 ± 0·2 in UP; again, this pattern was consistent across groups (Table 2): generalist predators – CropUP = 5·1 ± 0·3, UP = 0·03 ± 0·007; Homoptera – CropUP = 1·0 ± 0·2, UP = 0·005 ± 0·002; Heteroptera – CropUP = 0·2 ± 0·03, UP = 0·005 ± 0·002; Diptera – CropUP = 12·9 ± 0·4, UP = 0·1 ± 0·04; Linyphiidae – CropUP = 0·8 ± 0·1, UP = 0·003 ± 0·0006; Carabidae – CropUP = 0·03 ± 0·017, UP = 0·0005 ± 0·0003, UP = 0·0001 ± 0·0004; CFI – CropUP = 0·08 ± 0·02, UP = 0·001 ± 0·007, SFI – CropUP = 24·3 ± 0·4, UP = 0·4 ± 0·2. Invertebrate abundance within the crop fluctuated over the sampling period. There were time × treatment interactions for Staphylinidae, total Coleoptera and phytophages. However, the trend held and there were negligible numbers of these groups in UP at any time.

Table 4. anova table for analyses of crop-active invertebrate data collected by Dvac in 2002 and 2003. The table shows F statistics for treatment effects of individual undrilled patches (UP) /crop surrounding patches (CropUP) and the differences in these effects over the sampling period (time)
Source of variation20022003
Treatment
(UP/CropUP)
(d.f. = 1, 9)
TIme
(May/June/July)
(d.f. = 2, 34)
Treat*time
(d.f. = 2, 34 df)
Treatment
(UP/Crop)
(d.f. = 1, 8)
Time
(May/June/July)
(d.f. = 4, 32)
Treat*time
(d.f. = 2, 32)
  • *

    P < 0·05;

  • **

    P < 0·01;

  • ***

    P < 0·001; nsd, no significant difference.

Araneae
 Linyphiidae1·82nsd9·7**0·36nsd82·13***2·22nsd1·08nsd
Coleoptera
 Carabidae1·57nsd2·22nsd0·60nsd15·8**4·28*4·24*
 Staphylinidae1·52nsd2·82nsd0·20nsd23·55***8·45**8·73*
 Total Coleoptera0·32nsd1·04nsd0·73nsd38·07***2·61nsd0·84nsd
Diptera5·79*12·81***3·61nsd215·9***3·41nsd0·24nsd
Hemiptera
 Heteroptera9·18**4·38*3·84*7·64*1·15nsd1·12nsd
 Homoptera< 0·01nsd1·36nsd1·28nsd29·07***0·03nsd0·36nsd
Composite variables
 Generalist predators0·01nsd45·15***0·43nsd142·5***5·42**1·04nsd
 Phytophages3·64*17·16***0·17nsd76·07***11·62***4·82*
 Total invertebrates4·96*11·72***3·87*103·7***4·89*0·81nsd
 Skylark food items (SFI)2·17*1·42nsd0·12nsd15·25**4·72*0·76nsd
 Chick food index (CFI)4·94*11·45***3·96*102·1***1·32nsd1·52nsd

weed–invertebrate relationships

In 2002 (after fitting structural variables), ground-active phytophages were negatively associated with crop cover, Staphylinidae were positively associated with grass cover, crop cover and bare ground. Total ground-dwelling invertebrates, generalist predators and Carabidae did not show any relationship with vegetation cover. Crop-dwelling total invertebrates, generalist predators, phytophages, and SFI sampled by Dvac were positively associated with broadleaf cover (Table 5; for the full model, see Supporting Information, Table S1). The chick-food index (CFI) was positively associated with bare ground. In 2003, ground-active invertebrates were largely unrelated to vegetation cover; only Staphylinidae was positively associated with grass cover, crop cover and bare ground. Crop-dwelling total invertebrates, generalist predators, phytophages and skylark food items (SFI) sampled by Dvac were again positively associated with broadleaf cover (Table 5). In addition, total invertebrates and phytophages and SFI responded to crop cover and generalist predators were positively related to grass cover. In contrast to 2002, CFI was negatively associated with bare ground.

Table 5.  GLM model (quasi-Poisson) after first adding the structural variables which were: time of sampling (for Dvac data only), site/field and treatment. EXP, proportion of deviance explained by the model; *P ≤ 0·05; **P < 0·01; ***P < 0·001; only significant explanatory variables are shown here (see Supporting Information, Table S1 for full model)
Dvac 2002 (d.f. 25 = 609, n = 634)Dvac 2003 (d.f. = 25, 525, n = 550)
 βSEt βSEt
Total invertebrates (EXP: 59·6%)Total Invertebrates (EXP: 71·6%)
 Broadleaf0·0350·0103·69*** Broadleaf0·0450·00315·20***
     Crop0·0120·0052·25*
Generalist predators (EXP: 59·7%)Generalist predators (EXP: 33·4%)
 Broadleaf0·0260·0112·43* Broadleaf0·0190·0054·17***
     Grass0·0180·0072·65**
Phytophages (EXP: 84·7%)Phytophages (EXP: 85·43%)
 Broadleaf0·0760·0155·012*** Broadleaf0·0420·00313·78***
     Crop0·030·0064·94***
SFI (EXP: 60%)SFI (EXP: 72·47%)
 Broadleaf0·0360·013·71*** Broadleaf0·0450·00315·23***
     Crop0·0110·0052·65*
CFI (EXP: 50·6%)  CFI (EXP: 33·87%)  
 BG0·0040·0022·080* BG–0·0180·007–2·67**
     Litter<–0·0010·009–0·008nsd
Pitfall 2002 (d.f. = 25, 263, n = 287)Pitfall 2003 (d.f. = 24, 250, n = 274)
 βSEt βSEt
Phytophages (EXP: 61·8%)Phytophages (EXP: 76·9%)
 Crop–0·0140·006–2·48* Crop–0·0010·008–0·12nsd
Staphylinidae (EXP: 55%)Staphylinidae (EXP: 67·4%)
 Grass0·0360·0065·84***    
 Crop0·0120·0062·11*Crop0·0230·0073·49***
 BG0·0150·0072·21*BG0·030·0142·20*

As the GLM indicated a poor relationship between ground-active invertebrates and July vegetation cover, redundancy analysis was restricted to crop-active invertebrates. In 2002, site differences explained 8·5% of the variation. After the effect of site had been removed, the environmental variables explained 21·5% of the remaining variation (sum of all canonical eigenvalues 0·197) and the relationship was shown to be significant by Monte Carlo global permutation test of both axes (F = 18·9, P = 0·002). The sampling period May (F = 33·3, P = 0·002), treatment UP (F = 15·2, P = 0·002) and vegetation components grass cover (F = 8·6, P = 0·002), broadleaf cover (F = 88·8, P = 0·002) and bare ground (F = 7·7, P = 0·002) significantly influenced the invertebrate community structure (see Supporting Information, Table S2).

The triplot (Fig. 3) shows only taxa which explained > 5% of the variance as determined by CFit scores. Bare ground was positively correlated with environmental variables May (0·2024), WSR (0·0920) and CONV (0·0076) but negatively correlated with CropUP (–0·0102), grass cover (–0·6997), broadleaf cover (–0·2151), crop cover (–0·0351) and UP (–0·0898). Grass cover was positively correlated with July (0·2375), UP (0·4140) and, albeit weakly, with broadleaf cover (0·0816) and negatively with CONV (–0·1453), CropUP (–0·0697), WSR (–0·1976) and crop cover (–0·2295). Additionally, broadleaf cover was positively correlated with July (0·2442) and UP (0·3892), but negatively correlated with CONV (–0·1278), CropUP (–0·1082), WSR (–0·1520) and Crop (–0·2506).

Figure 3.

Redundancy analysis triplot for in-crop invertebrates collected by Dvac suction sampler in 2002. Eigenvalues: axis 1 = 0·147, axis 2 = 0·038. Bold line vectors indicate continuous environmental effects, nominal variables indicated by circles. Environmental variables: broadleaf cover (%), grass cover (%), crop cover (%), litter (%). May, early sampling period; July, late summer sampling period; CONV, conventional fields; UP, undrilled patches; CropUP, crop in fields with patches, WSR, wide-spaced row fields. Dashed line vectors represent invertebrate taxa.

In 2003, site differences explained 3·4% of the variation. After the effect of site had been removed, the environmental variables explained 23·6% of the remaining variation (sum of all canonical eigenvalues 0·228) and the relationship was shown to be significant by a Monte Carlo global permutation test of both axes (F = 18·3, P = 0·02). The sampling period May (F = 5·6, P = 0·002), treatment UP (F = 7·3, P = 0·002) and vegetation component broadleaf cover (F = 146·5, P =0·02) significantly influenced the invertebrate community structure (see Supporting Information, Table S2).

The triplot (Fig. 4) again only shows taxa which explained more > 5% of the variance as determined by CFit scores. Bare ground was positively correlated with May (0·5677) and UP (0·2374), but negatively correlated with CONV (–0·0791), CropUP (–0·0520), WSR (–0·1070), grass cover (–0·3112), broadleaf cover (–0·1577) and crop cover (–0·5723). Grass cover was positively correlated with UP (0·3189) and broadleaf cover (0·0428) and negatively correlated with May (–0·2861), CONV (–0·2163), CropUP (–0·0587) and WSR (–0·0395). Additionally, broadleaf cover was correlated with UP (0·1973) and CONV (0·0340) and negatively correlated with CropUP (–0·1656), WSR (–0·0653) and crop cover (–0·1541).

Figure 4.

Redundancy analysis triplot for in-crop invertebrates collected by Dvac suction sampler in 2003. Eigenvalues: axis 1 = 0·227, axis 2 = 0·001. Bold line vectors indicate continuous environmental effects, nominal variables indicated by circles. Environmental variables: broadleaf cover (%), grass cover (%), crop cover (%), litter (%). May, early sampling period; July, late summer sampling period; CONV, conventional fields; UP, undrilled patches; CropUP, crop in fields with patches; WSR, wide-spaced row fields. Dashed line vectors represent invertebrate taxa.

skylark diet

In 2002, Araneae, Diptera and Coleoptera were the most frequent prey classes, forming approximately 78% of the diet. Coleoptera was largely composed of Carabidae (adults + larvae) (69%), Staphylinidae (adults + larvae) (5%), Curculionidae (4%), Elateridae (4%), Chrysomelidae (0·8%), Nitidulidae (0·6%) and others (17%). Large carabids were the most frequently occurring prey and were recorded in 73% of the samples, followed by Tipulidae larvae (69%), adult Tipulidae (66%), other unclassified Diptera (67%) and Araneae (60%). Lepidoptera and Hymenoptera (mostly Symphyta and Ichneumonidae) together formed 8% of the diet and occurred in approximately 20% and 33% of the samples, respectively. Very little cereal was found in the faecal material, only three samples contained grain.

In 2003, Araneae, Diptera and Coeleoptera were still the most frequent prey classes but together formed only 58% of skylark diet, less than in the previous year. Coleoptera was composed of Carabidae (adults + larvae) (65%), Staphylinidae (adults + larvae) (9%), Curculionidae (9%), Elateridae (5%), Chrysomelidae (0·5%), Nitidulidae (0·2%) and others (3%). Large carabids were the most frequently occurring prey and were recorded in 76% of the samples, followed by Araneae (61%), cereals (50%), Tipulidae larvae (41%), adult Tipulidae (31%), and other unclassified Diptera (67%). Although cereal comprised only 7% of the diet, it occurred regularly in samples, unlike the previous year. Together, Lepidoptera and Hymenoptera formed 20% of the diet and occurred in 77% and 41% of the samples, respectively, an increase in relative proportion when compared with 2002.

Compositional analysis showed that there was no significant difference in dietary composition as shown by faecal material between years (F7,40 = 2·1, P = 0·062) or between treatments (F14,80 = 1·2, P = 0·26), however, there was a significant difference between sampling dates (F35,170 = 2·5, P≤ 0·001). The groups that differed were Araneae (F5,46 = 5·1, P = 0·001), Adult Hymenoptera (F5,46 = 3·4, P = 0·01), Carabidae, (F5,46 = 9·2, P ≤ 0·001), other Coleoptera (F5,46 = 11, P ≤ 0·001) and Diptera (F5,46 = 6·8, P ≤ 0·001).

Discussion

Incorporating undrilled patches in wheat crops has been shown to benefit skylark breeding success by increasing the number of chicks per breeding attempt by 0·5 (Morris et al. 2004), but the findings presented here highlight that this approach did not consistently increase the abundance of either invertebrate chick-food (using CFI and SFI as indicators) or invertebrates per se. The CFI results emphasize how poor winter cereals are as a resource because the CFI rarely rose above 0·2 and was frequently < 0·1 which is well below the level (0·7) considered necessary to maintain grey partridge populations (Potts & Aebischer 1995). This is a useful guide because some of the key groups are also important in the diet of other farmland bird chicks (Holland et al. 2006). Although SFI was related to broadleaf cover in both years, the CFI (a weighted index and more conservative measure) was poorly explained by analyses and showed a contradictory response to bare ground over 2 years. Given that the undrilled patches occupy just 0·3% of a sown field, it is unsurprising that the scope of the treatment did not consistently extend to the provision of resources for invertebrates at the field-scale.

Creating UP did affect the local distribution of some species. In 2002, more ground-active phytophages were found in UP when compared to the surrounding crop, as were crop-active phytophages, heteropteran species and a relatively high abundance of species selected by skylarks as chick-food; overall, chick-food may have been more available. This is consistent with foraging data which showed that within the UDPW treatment, UP were positively selected by foraging birds (Morris et al. 2007). In contrast, in 2003, UP remained bare of vegetation and were largely uncolonized by invertebrates. It is clear that the effectiveness of UP will be determined by large- and small-scale processes such as annual weather patterns and weed seed abundance and distribution respectively, and these will be difficult to predict. Further research investigating the establishment and subsequent management of UP could improve the consistency of the weed and invertebrate response.

It was expected that wider spaces between crop rows would allow colonization by weeds which would in turn support invertebrates. However, the redundancy analysis showed a negative relationship between WSR and both grass and broadleaf cover; overall abundance of invertebrates was not demonstrably increased by WSR.

Weed cover, particularly grass cover, increases the abundance of generalist predators such as ground beetles and spiders (Speight & Lawton 1976) and drives both primary and secondary producers in the invertebrate community (Bohan et al. 2007). However, in our study, ground-active invertebrates were not strongly related to vegetation cover. Possibly, vegetation cover was too low or perhaps, as we did not collect weed data in June and used July data as a surrogate, our analysis failed to detect the relationship. For crop-active invertebrates, broadleaf cover was the most consistently influential variable; in 2002, it was the only vegetation component related to four of the five invertebrate groups. In 2003, broadleaf cover remained important, although grass cover (predators) and crop cover (phytophages and SFI) were also were significant. Redundancy analysis supported the GLM results confirming that UP were likely to be colonized by broadleaved weeds and grasses and that these variables were positively correlated with the invertebrate community composition, broadleaf cover consistently so. It is likely that a certain threshold of cover must be reached before any significant effect on invertebrates is detected (Speight & Lawton 1976; Sotherton 1982), yet farmers may not tolerate weeds at densities needed to support invertebrates. Achieving a balance between controlling pernicious species whilst allowing beneficial species (at acceptable densities) to remain is challenging; using selective herbicides is one important management technique which can accomplish this. However, more research is needed to address concerns over potential yield loss and possible seed return which may increase the weed burden in following crops (Storkey & Westbury, 2007). The question remains whether the modest changes in invertebrate abundance/species assemblage brought about by undrilled patches had any impact on skylark diet; the impact of undrilled patches on invertebrate populations was low and skylark diet was not affected by treatment. That food was scarce in the crop in 2003 is borne out by the high incidence of cereal grain, not a preferred food for chicks, in the faecal samples. Furthermore, in 2003, species more commonly associated with grassland or wildflower strips such as Orthoptera (feeding on grasses and broadleaved weeds) and Neuroptera (feeding on nectar and pollen) were recorded, indicating that skylarks may have foraged in field margins. Forage watches carried out as part of this study confirm that the number of foraging flights to habitats outside the treatment areas was greater in 2003 (Morris et al. 2007). Skylarks take the most abundant prey and are not particularly specialized in their diet (Weibull 1999). Nevertheless, skylarks responded positively to the undrilled patches as demonstrated by their increased breeding success and we conjecture that the beneficial effect of UP for skylarks is linked to the provision of access points in the crop for foraging birds. A limited beneficial effect of WSR for skylarks was recorded by Morris et al. (2004) and was attributed to increased access; food accessibility is critical for skylarks (Odderskaer et al. 1997).

Clearly, there are obstacles to assessing the non-market value of agri-environment schemes or placing a value on agriculture's overall impact on biodiversity. However, it is important to ensure that the environmental value of agri-environment schemes should exceed or at least match actual payments (Glebe 2007). To improve the effectiveness of undrilled patches within a crop, we need to determine (i) the optimum number of patches per unit area; (ii) whether patches could produce sufficient invertebrates to effect skylark productivity and if this is so, (iii) develop techniques to more reliably create patches containing sufficient weed cover. The benefit of increasing in-field invertebrate numbers via undrilled patches is limited to the provision of food for birds. Other techniques to increase invertebrates across the whole field are needed if other ecological services, such as pest control and pollination, are to be improved.

Acknowledgements

The authors thank Nicholas Aebischer and Andrew Hoodless for statistical advice and two anonymous reviewers for their constructive comments. The work was part of the SAFFIE (LK0926) project. Our research partners in this work were: ADAS, RSPB, BTO, CSL and CAER, University of Reading. The project was sponsored by government departments, levy boards, wildlife and farming organizations, food retailers and the farming and crop protection industry's Voluntary Initiative, through the Sustainable Arable LINK programme.

Ancillary