The selection of stubble fields by wintering granivorous birds reflects vegetation cover and food abundance

Authors

  • D. Moorcroft,

    Corresponding author
    1. Edward Grey Institute for Ornithology, Department of Zoology, University of Oxford, South Parks Road, Oxford OX1 3PS, UK;
    2. RSPB, The Lodge, Sandy, Bedfordshire SG19 2 DL, UK; and
      D. Moorcroft, RSPB, The Lodge, Sandy, Bedfordshire SG19 2 DL, UK (fax +44 1767 683640; e-mail darren.moorcroft@rspb.org.uk).
    Search for more papers by this author
  • M. J. Whittingham,

    1. Edward Grey Institute for Ornithology, Department of Zoology, University of Oxford, South Parks Road, Oxford OX1 3PS, UK;
    Search for more papers by this author
  • R. B. Bradbury,

    1. Edward Grey Institute for Ornithology, Department of Zoology, University of Oxford, South Parks Road, Oxford OX1 3PS, UK;
    2. RSPB, The Lodge, Sandy, Bedfordshire SG19 2 DL, UK; and
    Search for more papers by this author
  • J. D. Wilson

    1. Edward Grey Institute for Ornithology, Department of Zoology, University of Oxford, South Parks Road, Oxford OX1 3PS, UK;
    2. RSPB Scotland, Dunedin House, 25 Ravelston Terrace, Edinburgh EH4 3TP, UK
    Search for more papers by this author

D. Moorcroft, RSPB, The Lodge, Sandy, Bedfordshire SG19 2 DL, UK (fax +44 1767 683640; e-mail darren.moorcroft@rspb.org.uk).

Summary

  • 1 Fields left fallow after harvest (i.e. stubble fields) support high wintering densities of many species of granivorous bird. We examined correlates of use by eight such species of different types of intensively managed wheat and barley stubble fields, organic wheat fields and set-aside fields on mixed lowland farmland in central England. Field occupancy was studied in relation to the physical characteristics of fields and seed abundance.
  • 2 Higher seed abundance was associated with greater occupancy by linnet Carduelis cannabina, grey partridge Perdix perdix, chaffinch Fringilla coelebs, yellowhammer Emberiza citrinella, reed bunting Emberiza schoeniclus and corn bunting Miliaria calandra. Larger areas of bare earth within stubble fields were associated with greater occupancy by linnet, yellowhammer, reed bunting and corn bunting, but lower occupancy by woodpigeon Columba palumbus.
  • 3 On conventional intensively farmed sites, seed abundance and area of bare earth were significantly greater on barley stubbles than on wheat stubbles.
  • 4 Seed numbers fell throughout the winter in all stubble types, although reductions were greatest on intensive barley stubbles, intermediate on intensive wheat stubble and lowest on undersown organic wheat stubbles.
  • 5 Within fields occupied by linnets, areas used for feeding had significantly greater quantities of seeds known to be important in their diet. Feeding areas also had a greater area of bare earth than randomly selected ‘non-feeding areas’.
  • 6 Linnets and reed buntings were rarely found on fields where densities of weed seeds important in their diets fell below 250 seeds m−2. In autumn, yellowhammers and grey partridges rarely fed on fields where cereal grain density was below 50 m−2. However, in spring, both species fed on these fields irrespective of grain density, perhaps indicating a switch to other food sources.
  • 7 We suggest that land managers wishing to maximize the value of overwinter stubble fields for granivorous birds locate such fields where there is a substantial natural regeneration of weed flora and where previous cropping (e.g. barley) is likely to offer a sparse stubble with substantial areas of bare ground.

Introduction

In recent decades, the populations of many farmland taxa have declined across Europe (e.g. plants: Rich & Woodruff 1996; invertebrates: Wilson et al. 1999; birds: Pain & Pienkowski 1997; mammals: Flowerdew 1997). They include several granivorous bird species of conservation concern (Gibbons et al. 1996) that rely on grain or weed seeds in the winter (Marchant & Gregory 1994; Wilson et al. 1999). A concomitant decline in their overwinter survival suggests that winter food availability could be involved (Siriwardena, Baille & Wilson 1998, 1999). The development of hardier varieties of autumn-sown cereals, technological advances in harvesting machinery and more effective weed control, have meant that the availability of overwinter stubble fields fell notably during the 1970s and 1980s (Chamberlain et al. 2000; Robinson & Sutherland 2002). Additionally, the quantity of seed on remaining fields has also declined markedly (Donald 1998; Robinson & Sutherland 2000; Watkinson et al. 2000). The population sizes of seed-eating birds may thus have been limited, particularly in winter when seeds are depleted over time (Robinson & Sutherland 2000) and because many granivorous passerines almost exclusively select stubble fields as winter foraging habitats (Bauer & Ranftl 1996; Wilson, Taylor & Muirhead 1996; Buckingham et al. 1999).

Stubble fields hold greater relative densities of seeds than other field types (Evans 1997; Robinson & Sutherland 2000), and their availability increased during the 1990s in the form of set-aside (Evans et al. 1998). It is therefore likely that the provision and management of overwinter stubbles, or alternatives such as game cover crops, will be an important component of agri-environment schemes designed to assist in reversing population declines of granivorous birds (Robinson, Wilson & Crick 2001). This study was designed to investigate how vegetation structure and seed availability on stubble fields influences variation in their use as foraging habitats by granivorous birds. We paid particular attention to within-field correlates of selection of foraging location by linnets Carduelis cannabina L.

Methods

STUDY SITES

This study was carried out from October to March in the winters of 1997–98 and 1998–99, on 122 stubble fields (40 in 1997–98 and 82 in 1989–99) chosen from 32 lowland farms in central England. All known barley Hordeum spp. and organic wheat Triticum spp. stubbles were included in order to maximize sample sizes. The remainder of fields were chosen randomly from a subset of fields expected to be stubble for the majority of the winter period. Other types of stubble fields (e.g. oilseed rape Brassica napus oliferia L. and oats Avena fatua L.) were very rare in the study region and were therefore excluded. All farms were typical of the local mixed farming regime, with varying proportions of land in arable production (cereals, broad-leaved crops) and pastoral agriculture. Seventeen stubble fields (from four farms, total area 165 ha) were managed organically for wheat, with the preceding crop undersown with mixtures of ryegrass Lolium spp. and white clover Trifolium repens L. Intensive wheat (n = 67, total area 802 ha) and intensive barley (n = 38, total area 409 ha) fields were managed in a conventional ‘intensive’ manner, having received routine inorganic fertilizer and pesticide applications. Typical pesticide applications for wheat or barley fields sown in the winter comprise approximately one insecticide, five fungicide and four herbicide sprays per rotation (Thomas, Garthwaite & Banham 1996). Each field examined was either overwintering as stubble in preparation for spring sowing of a cereal crop (OWS) or had been entered into the first year of a set-aside scheme (rotational set-aside, RSA; Evans et al. 1998). All RSA fields were also stubbles with a naturally regenerated vegetation cover and received no herbicide application during the study, in accordance with the Arable Area Payments and Set-aside Schemes (Evans et al. 1998). Prior to ploughing, OWS fields were typically sprayed with glyphosate, a non-selective, non-residual post-emergence herbicide that effectively controls grass and broad-leaved weeds (Worthing & Hance 1991). All organic wheat fields were OWS.

The areas of fields were either obtained directly from the farmer or calculated from 1 : 25 000 maps using an AutoCAD software package and a CalComp digitizing slate. The boundary characteristics of each field were recorded in order to calculate a boundary height index using the methods of Wilson et al. (1997). The perimeter of each field was divided into sections, according to the following categories: 0, no vertical structure; 1, a low (< 2 m) hedgerow, wall or bank without trees; 2, a tall (> 2 m) hedgerow, wall or bank without trees; 3, a hedgerow with trees or a line of trees; 4, woodland edge or other boundary type such as a garden, scrub or farm buildings. The length of each section (m) was multiplied by its category score, and the sum over all sections divided by the field perimeter to give a boundary height index.

BIRD COUNTS

Fields were visited monthly, when parallel transects (30 m apart) were walked along the fields’ longest axis in order to flush all birds on the field (Wilson, Taylor & Muirhead 1996; Perkins et al. 2000). All birds flushed in this manner were recorded. Care was taken to avoid double counting through observation of movements of previously flushed individuals. Censusing was only carried out on fields while they remained as stubble. If fields were ploughed during the period of the study, counts were no longer undertaken. Censusing periods avoided wet and windy weather, which is known to affect bird activity. All counts began 1 h after dawn and were completed at least 1 h before dusk, to avoid biases caused by birds leaving or arriving at roosting sites. The order in which the three observers visited stubble types was randomized in order to avoid any biases in diurnal variation within our survey periods.

FIELD CHARACTERISTICS

Populations of potential food resources (food plants and seeds) were sampled in two ways. First, quadrat surveys of plants were undertaken in October on all 122 fields. Forty quadrats (20 × 20 cm) were placed equidistantly along a diagonal transect across each field, and every plant species was given a percentage cover score relating to the total area of the quadrat (Bullock 1996). The percentage area occupied by bare soil was also recorded. Secondly, seed densities were recorded in a subsample of 27 fields (10 intensive wheat, 10 barley fields and all seven organic stubbles) chosen at random from the 40 fields sampled in the first winter. Sampling in the last 2 weeks of October 1997 and March 1998 allowed assessment of seed depletion over the winter. In each field, seeds were collected from 10 soil cores (15 cm diameter) collected at equidistant points along a randomly assigned diagonal transect. Only soil on or immediately below the surface (depth of 3 mm) was sampled in order to restrict sampling to that part of the soil seed bank likely to be available to small passerines (Green 1978; Robinson 1997). All ripe seeds present on standing plants within the sampling area of the core were also collected, although kept separate in order not to bias the measure of seed availability for those species unable to feed from the standing crop. Samples were placed in resealable polythene bags and stored within 24 h at 4 °C to prevent germination. This sampling regime represented a compromise between adequate sampling per field, number of field types and logistical constraints of seed bank analysis (Benoit, Kenkel & Cavers 1989) while providing a standardized measure of seed abundance. The seeds from each soil core were extracted by washing samples through sieves of decreasing mesh size (1 mm, 500 µm, 63 µm). Seeds were then allowed to dry and were individually identified using an appropriate guide (Flood & Gates 1986) and reference material collected in the field.

The characteristics of the three main stubble field types were compared using analysis of variance and Kruskal–Wallis tests. Where appropriate, we used post-hoc contrasts to investigate the main sources of variation between stubble types. Factor levels were grouped and then the model refitted with just two categories (using Mann–Whitney or t-tests, where appropriate). Such contrasting is more robust statistically than multiple comparisons, which lead to type I error (Crawley 1993; M. Crawley, personal communication). Mean percentage cover scores of botanical species and the area of bare earth per field were calculated over all quadrats within fields. All vegetation was identified to the species level, except Polygonaceae and Brassica spp. Seed densities were pooled across fields, as samples within fields cannot be considered independent. Knowledge of bird species diet was used to identify key dietary weed/seed species (henceforth referred to as dietary weeds or dietary seeds), Wilson et al. (1999).

RELATIONSHIPS BETWEEN BIRD USAGE AND FIELD CHARACTERISTICS

The relationships between bird usage and the characteristics of all 122 fields were quantified using logistic regression in GLIM 4.0 (Numerical Algorithms Group 1993). The number of occasions on which a species was present was specified as the response variable, and the total number of visits to that field (six) as the binomial denominator (specifying a binomial error structure and a logit link function; Crawley 1993). This approach tackles two problems inherent in using the original count data. First, the statistical non-independence of individuals in flocks is eliminated (Krebs 1987). Secondly, counts of individuals per field are often skewed and cannot easily be fitted by Poisson or negative binomial error distributions (Perkins et al. 2000). However, this approach does not eliminate the bias of birds in a flock in one field affecting the probability of occurrence of birds in neighbouring fields. Analyses were conducted for species for which (i) at least 50 individuals were recorded during the study; (ii) the species was recorded on at least 15 fields; and (iii) the frequency of occurrence on fields was significantly (P < 0·05) positively correlated with total count, indicating that significant explanatory predictors of field occupancy also explained significant variation in the numbers of individuals in occupied fields (Perkins et al. 2000). Eight species met these criteria: grey partridge Perdix perdix L., woodpigeon Columba palumbus L., skylark Alauda arvensis L., linnet, chaffinch Fringilla coelebs L., reed bunting Emberiza schoeniclus L., yellowhammer E. citrinella L. and corn bunting Miliaria calandra L. Logistic regression models were also developed for the subset of 27 fields on which seed densities were measured and where the species included in the models were predominantly or wholly feeding on a seed-based diet and encountered in 10 or more of the 27 fields.

‘Full’ models were constructed for each species using all predictor variables (Table 1). Quadratic terms for continuous predictors were included to account for curvilinear effects. The significance of predictors was assessed using the change in deviance (ΔD), which is distributed asymptotically as χ2 on removal of each predictor. In order to control for variation resulting in unmeasured parameters that may be site-specific, the factor ‘Farm’ was retained such that the null models were ‘1 + Farm’. Models were simplified using a backward selection procedure (Crawley 1993). The minimum adequate model (MAM) was identified when no predictors could be added or deleted from the model without causing a significant change in deviance (P < 0·05). Potential intercorrelations between predictors were examined using Spearman rank correlations, Mann–Whitney or Kruskal–Wallis tests as appropriate (see Appendices 1a and 1b). If a predictor excluded from the MAM was intercorrelated (P < 0·05) with another predictor retained in a MAM, alternative MAMs were constructed. Our modelling approach allowed a two-step process. First, we could test for the effect of crop type as an obvious and easily manipulated predictor. Secondly, by replacing crop type with correlated predictors describing bare earth and seed abundance, we could examine why crop type was important and how individual crops might be managed.

Table 1.  Predictor variables entered into the logistic regression modelling of field occupancy across (a) 122 fields at 32 lowland farm study sites and (b) 27 fields in which seeds were sampled
PredictorFactor/covariateCategory definition
  • *

    Dietary information for each species taken from Wilson et al. (1999).

  • Count includes seeds taken from the standing crop within the sampling area for only those species known to be capable of exploiting such resources.

(a) All 122 fields
YearFactor1: 1997–98; 2: 1998–99
FarmFactorFarm-specific code (1–32)
Preceding cropFactor1: intensive wheat; 2: intensive barley; 3: organic wheat
Field areaCo-variateMeasured in hectares
Stubble managementFactor0: stubble; 1: set-aside
Boundary indexCo-variateCalculated measure of boundary structure
Area of bare earthCo-variatePercentage cover of bare soil per field calculated over all quadrats
Number all weed species presentCo-variateNumber of non-crop plant species present
% weed coverCo-variatePercentage cover of all weed species per field calculated over all quadrats
Number diet weed species*Co-variateNumber of non-crop species that are known dietary components of particular species/suite of species per field calculated over all quadrats
% weed cover (dietary species)*Co-variatePercentage cover of weed species that are known dietary components of particular species/suite of species per field calculated over all quadrats
% volunteer cereal coverCo-variatePercentage cover of volunteer cereal plants per field calculated over all quadrats
(b) 27 fields in which seed populations were sampled
FarmFactorFarm-specific code (1–18)
Preceding cropFactor1: intensive wheat; 2: intensive barley; 3: organic wheat
Field areaCo-variateMeasured in hectares
Stubble managementFactor0: stubble; 1: set-aside
Boundary indexCo-variateCalculated measure of boundary structure
Total number weed seed speciesCo-variateNumber of weed seed species present per field within 10 soil cores
Total number seedsCo-variateNumber of all seeds present per field within 10 soil cores
Total number weed seedsCo-variateNumber of non-crop seeds present per field within 10 soil cores
Dietary weed-seed species abundance*†Co-variateNumber of non-crop species that are known dietary components of particular species/suite of species per field within 10 soil cores
Dietary weed-seed abundance*†Co-variateTotal number of seeds that are known dietary components of particular species/suite of species per field within 10 soil cores
Cereal grain abundanceCo-variateTotal number of cereal grains per field within 10 soil cores

The explanatory power of each model was assessed using the ratio of the residual deviance to the residual degrees of freedom. Ratios close to one suggest that the model provides a good fit to the data (Crawley 1993), whereas ratios above two indicate weak predictive power. To correct for overdispersion in models with ratios over two, standard errors were multiplied by √(residual deviance/residual degrees of freedom) (Crawley 1993).

We calculated changes in bird density with food abundance using Spearman rank correlations. Such an approach was not possible for our main modelling because we wished to investigate the effects of many predictors.

CHARACTERISTICS OF LINNET FORAGING LOCATIONS

To assess whether linnets were selecting foraging locations within fields on the basis of stubble characteristics and seed availability, plants and soil cores were sampled in five quadrats positioned along the major axis of any linnet flock whose position (henceforth termed ‘feeding’ area) could be accurately ascertained by observing feeding linnets and fresh faeces. Five further quadrats and soil cores were collected from a randomly located transect of equal length where linnets were not feeding in the same field (henceforth termed ‘non-feeding’ area). There was no overlap between feeding and non-feeding areas. The starting point of the non-feeding area transect was identified by random placement of the quadrat with further samples taken along the same directional axis as the feeding area transect. The same data set as described above was collected from these quadrats.

Soil cores collected, respectively, from feeding and non-feeding areas within each field were pooled, as they could not be considered independent. Wilcoxon paired-sample tests were used to compare stubble characteristics and seed availability of the two areas within a given stubble field.

Results

BIRD COUNTS

A total of 13 seed-eating species (21 853 individual records) was noted across all stubble fields, over a total of 667 bird-count field visits, but only eight species were analysed further.

FIELD CHARACTERISTICS

Field size ranged from 2·5 ha to 53·8 ha (mean ± SE = 11·3 ± 0·71) and did not differ significantly with preceding crop or stubble management (OWS vs. RSA: F1,120 = 0·07; NS), although it was significantly greater in 1998–99 (year: F1,120 = 4·81; P < 0·05). Boundary structure did not differ significantly with preceding crop type. Botanical composition (mean percentage cover) varied between stubble types (Table 2): by comparison with other locations, volunteer cereals had significantly greater cover in intensively managed wheat stubbles; clover and grass had significantly greater cover in undersown organic stubble fields; broad-leaved weeds had significantly greater cover in intensive barley stubbles. The area of bare earth (range 4·6–68·5%) was significantly greater on barley stubbles than on either wheat stubbles.

Table 2.  Characteristics of all three stubble types included in this study. Mean values (± 1 SE) are given
 Preceding cropSignificance*Post-hoc contrasting
 Intensive wheatIntensive barleyUndersown organic wheat
  • *

    Significance levels were tested by either Kruskal–Wallis tests with 2 degrees of freedom (H-values) or analysis of variance as appropriate. Where appropriate post-hoc contrasting was undertaken.

Number of fields673817  
Field area (ha)11·98 ± 0·93 10·77 ± 1·50 9·70 ± 1·23H2 = 1·87; NS 
Boundary index1·715 ± 0·082 1·547 ± 0·1501·714 ± 0·173H2 = 0·58; NS 
Number of weed species12·46 ± 0·372 18·18 ± 0·61712·65 ± 0·870F2,119 = 37·16; P < 0·001t60 = 8·05; P < 0·000 Barley vs. wheat
Percentage bare soil27·70 ± 1·44 37·52 ± 2·2720·61 ± 2·59H2 = 20·80; P < 0·001U84,38 = 4·31; P < 0·001 Barley vs. wheat
Percentage cover (volunteer cereal)33·69 ± 1·22 28·22 ± 1·3829·12 ± 2·19H2 = 9·60; P = 0·013U67,55 = 2·18; P = 0·018 Intensive wheat vs. barley/organic wheat
Percentage cover (grass) 5·88 ± 0·67  3·73 ± 0·399·88 ± 1·44H2 = 18·29; P < 0·001U105,17 = 6035·0; P < 0·001 Intensive vs. organic
Percentage cover (clover) 0·278 ± 0·0850·0033 ± 0·003343·60 ± 2·64H2 = 70·84; P < 0·001U105,17 = 5565·0; P < 0·001 Intensive vs. Organic
Percentage cover (broad-leaved weeds)39·64 ± 1·39 50·68 ± 2·1427·83 ± 2·91H2 = 32·43; P < 0·001U84,38 = 4288·5; P < 0·001 Barley vs. wheat

Of a total of 35 non-crop plant species encountered, only 16 species had greater than 1% of field cover in one or more stubble type. Each stubble field was typically dominated by a few food plant species. Meadowgrasses Poa spp. and chickweed Stellaria media L. were most common on undersown organic wheat stubbles, while black bindweed Fallopia convolvulus L., fat-hen Chenopodium album L. and field speedwell Veronica persica Poir. were the most common broad-leaved weeds on intensive stubbles. Barley stubbles had significantly greater mean percentage cover of broad-leaved weeds and weed species richness than both types of wheat stubbles (Table 2).

On the subset of 27 fields in which seed availability was measured, a total of 21 non-crop species was recorded in the seed flora, 12 of which were represented on all three stubble types. Seed densities per field in October (range 266·0–860·4 m−2), were significantly greater on intensive barley and organic wheat stubbles than intensive wheat stubbles (Table 3). Seed densities per field in March (range 220·8–747·2 m−2) were highest on undersown organic stubbles, intermediate on intensive barley stubbles and lowest on intensive wheat stubbles. Seed densities were lower across all stubble types compared with October densities (all species: mean percentage change −22·31 ± 3·62; range −49·7 to +3·1; n= 23: Table 3). Reductions in weed seed density differed significantly among stubble types, being greatest on intensive barley stubbles, intermediate on intensive wheat stubbles and lowest on undersown organic wheat stubbles. There was no significant difference in the mean number of seed species found on each stubble type in either sampling month (Table 3).

Table 3.  Densities of weed seeds (m−2) and number of species found across 27 stubble fields at 18 farm study sites in central England. Mean values (± SE) are given. †Significance levels were tested by Kruskal–Wallis tests with 2 degrees of freedom: * P < 0·05; ** P < 0·01; *** P < 0·001; NS, not significant. Post-hoc contrasting was used, where appropriate
 Intensive wheatIntensive barleyUndersown organic wheatSignificancePost-hoc contrasting
October
Number of fields10107  
Seed density m−2 434·1 ± 25·9 647·0 ± 44·1 614·6 ± 63·9H = 11·26; P = 0·004U17,10 = 74·0; P = 0·001 Intensive barley/organic wheat > intensive wheat
Number of seed species   3·0 ± 0·365   3·8 ± 0·442  3·29 ± 0·522H = 1·84; P = 0·404; NS 
March
Number of fields1067  
Seed density m−2 330·6 ± 25·7 449·1 ± 52·1 541·8 ± 59·0H = 10·30; P = 0·006U13,10 = 70·0; P = 0·002 Intensive barley/organic wheat > intensive wheat
Number of seed species    2·8 ± 0·249   3·7 ± 0·651  3·86 ± 0·595H = 2·57; P = 0·288; NS 
Mean % change (overall seed density)−22·68 ± 5·39−34·64 ± 6·86−11·21 ± 4·46H = 6·51; P = 0·031*U17,9 = 239·0; P = 0·016 Barley > wheat

RELATIONSHIPS BETWEEN BIRD OCCUPANCY AND FIELD CHARACTERISTICS

On all 122 fields, farm was the strongest predictor of bird occupancy in all models (Table 4). Those predictor variables retained in MAMs both with and without the inclusion of farm could be considered to explain variation in the frequency of occurrence of birds on fields both within and between farms. Ratios of the residual deviance to the residual degrees of freedom were for most models between one and two, indicated generally good fit to the data.

Table 4.  Results of logistic regression analyses for eight bird species surveyed on 122 stubble fields at 32 lowland farms in central England. Direction of association (+ or –) between predictor variables and frequency of occurrence (0–6) are given for each MAM. Note that significant predictors are given in bold while those not retained when farm was added to the model are given in italics. Models 1–3 are alternative final MAM for each species carried out because some predictors were intercorrelated (see text for further explanation)
 Model goodness- of-fit (res. dev./res. d.f.)Year 1 = 1997–98 2 = 1998−99Preceding crop 1 = intensive wheat 2 = intensive barley 3 = organic wheatField areaBoundary indexStubble management stubble (0)/ set-aside (1)Area of bare earthNumber of all weed species present% weed coverNumber of diet weed species% weed cover (dietary species)% volunteer cereal cover
  1. res. dev., residual deviance; res. d.f., residual degrees of freedom.

Linnet (Model 1)1·650·0770·000 (2 > 1 = 3)0·000 (+)0·7400·0630·6740·0830·0850·1480·7100·786
Linnet (Model 2)1·530·3510·3450·000 (+)0·6040·040 (0 > 1)0·000 (+)0·7720·5180·8540·000 (+)0·226
Linnet (Model 3)2·430·1200·2020·000 (+)0·2020·2440·0600·002 (+)0·3720·0940·0630·938
Grey partridge1·650·022 (1 > 2)0·1840·7930·2820·1370·0770·5440·0830·5950·000 (+)0·830
Woodpigeon (Model 1)1·750·1470·000 (3 > 1 > 2)0·4540·0650·000 (0 > 1)0·0630·2910·0760·0540·0770·134
Woodpigeon (Model 2)2·730·9310·1560·1990·9900·034 (0 > 1)0·000 (–)0·1190·3880·1440·8740·233
Skylark (Model 1)1·610·3470·000 (2 > 1 > 3)0·000 (+)0·5310·5720·8620·1510·1670·3650·0850·213
Skylark (Model 2)3·250·6170·1510·000 (+)0·6980·0680·0630·1660·0650·9250·046 (+)0·065
Chaffinch0·980·0770·8460·5600·0770·7710·5700·4950·9890·7960·000 (+)0·057
Yellowhammer (Model 1)2·260·6450·7600·0810·7010·6880·002 (+)0·6520·9330·4220·000 (+)0·071
Yellowhammer (Model 2)2·770·1940·040 (2 > 1 > 3)0·4400·6620·5470·0600·2200·4500·1060·0970·169
Reed bunting (Model 1)2·140·011 (1 > 2)0·000 (2 > 1 = 3)0·0700·0910·0840·0670·5670·6150·6110·0570·969
Reed bunting (Model 2)1·790·007 (1 > 2)0·9750·5420·0610·037 (0 > 1)0·000 (+)0·1730·1970·1760·000 (+)0·084
Corn bunting (Model 1)3·070·0750·000 (2 > 1 = 3)0·2780·1260·6700·0760·2090·1090·9060·0820·156
Corn bunting (Model 2)1·560·2130·9750·2050·2230·3070·000 (+)0·2290·0810·9880·000 (+)0·014 (+)

Stubble field occupancy by two species (skylark and linnet) increased with field area, possibly reflecting either active selection by a species for larger fields or increased probability of field occupancy on larger fields if birds were distributed randomly.

Field occupancy by five species (woodpigeon, skylark, linnet, reed bunting and corn bunting) appeared to be significantly influenced by the preceding crop (Table 4 and Fig. 1). Yellowhammer occupancy was also influenced in this way, but only if the effect of farm was not controlled for. Field occupancy was significantly greater on intensive barley stubbles for all these species except woodpigeon, which was significantly greater on undersown organic wheat stubbles. No species was most strongly associated with intensive wheat stubbles. Due to significant intercorrelation between preceding crop type and percentage cover of dietary weeds and bare earth, alternative MAMs were constructed. Percentage bare earth was retained for woodpigeon (negative relationship), while both variables were retained for linnet, reed bunting and corn bunting (positive relationship) (Table 4). Thus, both percentage cover variables were strong predictors of field occupancy for these species, and may be important in explaining differences in stubble type occupancy. Grey partridge, chaffinch and yellowhammer were positively associated with the percentage cover of dietary weeds independently of preceding crop type. Yellowhammer occupancy was also positively associated with area of bare earth independently of preceding crop type (Fig. 2).

Figure 1.

Mean occupancy of three types of stubble fields (% of visits) for eight granivorous species in central England.

Figure 2.

(a) The predicted effect of the proportion of bare earth on the mean occupancy of each field. Probability of field use was derived by solving the equations of the MAMs for each species (those with farm included, see Table 4). The value of the plotted predictor (in this case bare earth) was varied within the bounds it was recorded in the field and substituted into the MAM equation while holding constant the values of the other predictors. For continuous variables, these were set to the mean value of all 122 fields while factors were set to level 1 (e.g. farm 1, year 1, etc.). Note that the intercept of a plot will vary according to the values at which the variables held constant in the MAM are set, dependent on the species relationship with these variables, but the gradient of the slope will remain the same. (b) The predicted effect of the proportion of weed cover of known food plants for a given species on the mean occupancy of each field by that species. For explanation of derivation of lines see (a). Note that the proportion of weed cover was only varied within the bounds that it was recorded in the field for each species.

Sufficient data for six species (grey partridge, skylark, linnet, yellowhammer, reed bunting and corn bunting) allowed analysis of the effects of seed resources. October seed density was used because ploughing prior to March meant seed density in some fields could not be measured. For all species, farm was the strongest statistical predictor of occurrence. Other significant positive predictors were area of bare earth (linnet and corn bunting), cereal grain abundance (grey partridge and yellowhammer) and dietary weed seed abundance (linnet and reed bunting) (Table 5).

Table 5.  Results of logistic regression analyses for six bird species surveyed on 27 stubble fields at 18 lowland farms in central England where seed resources were investigated. Direction of association (+ or −) between predictor variables and frequency of occurrence (0–6) are given for each MAM. Note that significant predictors are given in bold while those not retained when farm was added to the model are given in italics
 Model goodness-of-fit (res. dev./res. d.f.)Preceding crop 1 = intensive wheat 2 = intensive barley 3 = organic wheatField areaStubble management stubble (0)/ set-aside (1)Boundary indexArea of bare earthTotal number weed seed speciesTotaln umber seedsTotal number weed seedsDietary weed seed species abundanceDietary weed seed abundanceCereal grain abundance
Linnet0·870·5890·2880·2170·7600·040 (+)0·430·0650·0820·0550·000 (+)0·894
Grey partridge1·130·5120·005 (+)0·9860·1980·3940·6440·5460·4520·2400·6720·000 (+)
Skylark4·000·000 (2 > 1 > 3)0·004 (+)0·006 (0 > 1)0·7750·3200·4740·2820·2630·3980·4600·117
Yellowhammer2·100·4730·012 (+)0·1380·7790·2020·3310·1630·7010·5050·3450·000 (+)
Reed bunting2·220·1470·5430·0620·0930·9210·1590·2080·2110·1460·000 (+)0·741
Corn bunting1·650·0970·3660·7480·0830·000 (+)0·1150·4100·4620·0600·9560·002 (+)

The density of four species increased significantly with key dietary seed density: linnet (dietary seeds in both October and March on intensive stubble fields, although not on organic stubbles), yellowhammer (cereal grain within individual stubble types in October, although not in March), grey partridge (cereal grain on intensive wheat stubbles) and reed bunting (dietary weed seeds in both October and March across all stubble types) (Table 6a).

Table 6a.  Relationships between the density of four granivorous bird species and weed seeds known to be important in the diet of each species (determined using Wilson et al. 1997) on 27 stubble fields in October 1997 and on 23 fields in March 1998. Spearman rank correlation coefficients (rs) and their significance values are presented below. Results significant at P < 0·05 are given in bold. As an example, the density of linnets was significantly related to the density of seeds (known to be important in the diet of linnets) in October on 10 intensive wheat stubbles at P = 0·01. A single field is equivalent to one replicate
SpeciesCrop typeAll crop types
Intensive wheat (October) n = 10 fieldsIntensive wheat (March) n = 10 fieldsIntensive barley (October) n = 10 fieldsIntensive barley (March) n = 6 fieldsOrganic undersown wheat (October) n =  7 fieldsOrganic undersown wheat (March) n = 7 fieldsTotal using all fields (October) n = 27 fieldsTotal using all fields (March) n = 23 fields
Linnetrs = 0·692 (P = 0·01)rs = 0·72 (P = 0·005)rs = 0·928 (P < 0·001)rs = 0·812 (P = 0·002)rs = 0·427 (P = 0·339)rs = 0·335 (P = 0·335)rs = 0·549 (P = 0·002)rs = 0·583 (P = 0·002)
Yellowhammerrs = 0·634 (P = 0·03)rs = 0·377 (P = 0·282)rs = 0·659 (P = 0·02)rs = 0·625 (P = 0·095)rs = 0·755 (P = 0·02)rs = 0·069 (P = 0·884)rs = 0·621 (P < 0·001)rs = 0·386 (P = 0·069)
Grey partridgers = 0·942 (P < 0·001)rs = 0·462 (P = 0·179)rs = 0·393 (P = 0·261)rs = 0·348 (P = 0·449)rs = 0·112 (P = 0·81)rs = 0·16 (P = 0·731)rs = 0·517 (P = 0·004)rs = 0·292 (P = 0·176)
Reed buntingrs = 0·7 (P = 0·012)rs = 0·195 (P = 0·589)rs = 0·804 (P = 0·001)rs = 0·928 (P < 0·001)rs = 0·267 (P = 0·562)rs = 0·111 (P = 0·812)rs = 0·509 (P = 0·005)rs = 0·507 (P = 0·011)

Changes in density, between the start and end of winter, were also significantly positively correlated with changes in key dietary seed density for three species: linnet (dietary seeds across all stubble types and within intensive barley stubbles), grey partridge (cereal grain across all stubble types and within intensive wheat stubbles) and reed bunting (dietary seeds across all stubble types, but not on undersown organic wheat stubbles) (Table 6b).

Table 6b.  Relationship between the change in density of four granivorous bird species and weed seeds known to be important in the diet of each species (determined using Wilson et al. 1997) on 23 stubble fields in central England between October 1997 and March 1998. Spearman rank correlation coefficients (rs) and their significance values are presented below. Results significant at P < 0·05 are given in bold. A single field is equivalent to one replicate
SpeciesCrop typeAll crop types
Intensive barley (between October and March) n= 10 fieldsIntensive wheat (between October and March) n= 10 fieldsOrganic undersown wheat (between October and March) n= 7 fieldsTotal using all fields (between October and March) n= 23 fields
Linnetrs= 0·335 (P= 0·343)rs= 0·928 (P < 0·001)rs= 0·089 (P= 0·849)rs= 0·559 (P= 0·004)
Yellowhammerrs= 0·41 (P= 0·293)rs= 0·091 (P= 0·864)rs=−0·135 (P= 0·773)rs= 0·109 (P= 0·62)
Grey partridgers= 0·794 (P= 0·002)rs= 0·543 (P= 0·266)rs=−0·229 (P= 0·622)rs= 0·51 (P= 0·012)
Reed buntingrs= 0·64 (P= 0·025)rs= 0·829 (P= 0·014)rs=−0·075 (P= 0·873)rs= 0·646 (P < 0·001)

LINNET FORAGING SITE SELECTION

The foraging locations of linnet flocks were identified in 11 stubble fields (three intensive wheat, six intensive barley, two undersown organic wheat) in 1997–98. Feeding areas held a significantly greater density of dietary weed seeds (mean ± SE = 302·9 ± 49·3 m−2) than non-feeding areas (mean ± SE = 165·9 ± 19·2 m−2; Wilcoxon paired-sample test: U11,11 = 63·0; P= 0·009) and greater mean percentage bare earth (feeding area: mean ± SE = 40·84 ± 2·84%; non-feeding area: mean ± SE = 32·78 ± 2·78%; Wilcoxon paired-sample test: U11,11 = 65·0; P= 0·005). There was no significant difference between feeding and non-feeding areas in the species richness of weed seeds (P = 0·790; NS), all seeds (P = 0·722; NS), dietary weed seeds (P = 0·285; NS) or percentage weed cover scores (P = 0·307; NS).

Discussion

This study has highlighted how variation in the abundance and availability of weed and seed food resources between field types might affect the distribution of wintering seed-eating species between and within stubble fields. Linnet distribution, both across and within intensively managed stubble fields, was strongly correlated with food resources. The lack of a significant relationship within undersown stubble fields probably reflects the difference between food resource abundance and availability (Hutto 1990). Undersowing greatly increases ground cover and reduces the incidence of weed plants (Fisher & Davies 1991), presumably through limited germination opportunities (Lawson et al. 1992). Therefore, while organic stubbles may hold comparable or higher seed densities than conventional stubbles, as a result of previous management (Samuel & Guest 1990; Moreby et al. 1994; Albrecht & Forster 1996), food availability to the birds may be significantly lower than in more open conditions.

Seed availability may have been approximated by area of bare earth on stubble fields in this study, and positive association with the occupancy of several seed-eating birds accords with others (Grzybowski 1983; Clarke et al. 1997; Perkins et al. 2000). Woodpigeon occupancy, however, declined with increasing bare earth, presumably because it grazes on developing vegetation in winter, particularly clover Trifolium repens L. (Murton, Isaacson & Westwood 1966) and oilseed rape (Inglis et al. 1997), the former of which may explain its greater occupancy of undersown stubble fields. No relationship between skylark occupancy and bare earth is perhaps because of its diverse diet, including leaf material, cereal grain and weed seeds (Robinson 1997; Wilson et al. 1999).

For seed-eating passerines such as linnet, yellowhammer, reed bunting and corn bunting, which do not graze vegetation but whose winter diet consists almost exclusively of seeds, greater access to the bare soil may increase foraging efficiency by reducing the physiological costs of foraging within dense vegetation, which is both wetter and harder to move in (Clarke et al. 1997). In winter, dew and frost often lead to cold and wet vegetation, prolonged exposure to which can increase the energetic demands on birds (Dawson, Carey & Van’t Hof 1992). Bare soil will also increase the detectability of seeds (Whitehead, Wright & Cotton 1995; Robinson 1997) and allow greater efficiency of predator detection (Grzybowski 1983; Metcalfe 1984).

Buckingham et al. (1999) noted that barley stubbles had relatively higher bird numbers than other crop types. While supported by this study, greater field occupancy by many seed-eating species increased with areas of bare earth and greater food resources, both being strongly correlated with barley stubbles. Thus, the value of barley stubbles probably results from the crop being sparser and weedier than wheat stubbles. Barley has a short straw and, due to an early harvest, does not overshadow potentially establishing weed species (Fitter & Ashmore 1974; Soffe 1995). Lower vegetation density can also permit greater weed seed production (Wright 1993) as well as permitting bare areas to remain beneath the crop (Soffe 1995). Additionally, barley crops typically receive fewer herbicide treatments than wheat crops (Thomas, Garthwaite & Banham 1996) and therefore hold greater numbers of weed seeds (Don 1997).

Linnets and reed buntings were rarely found on stubble fields when they held densities of dietary weed seeds below 250 seeds m−2, although many fields with seed densities above this threshold could be unoccupied on any one visit. Additionally, within fields, linnets fed in patches that held mean dietary seed densities of 300 seeds m−2 or more. Thus, depletion of seed resources may reduce field carrying capacity. In this study, the densities of dietary seeds fell below 250 seeds m−2 by March for linnet and reed bunting in around half the fields sampled (12 and 14 out of 23 fields, respectively). In contrast, the number of fields below this density in October was only 22% for linnet and 37% for reed bunting.

Seed depletion was greatest on intensive barley fields compared with other stubble types, possibly as a result of greater predation, germination or rotting of seeds on barley stubbles. While a higher incidence of seed rotting is unlikely in the more open sward associated with barley stubbles, such a sward would facilitate greater germination (Wright 1993). Additionally, while the effects of other seed predators in the UK (e.g. insects and mammals) are negligible (Robinson 1997; but see Tellaria, Santos & Diaz 1994), greater associated densities of seed-eating passerines might increase depletion. Seed resources on winter stubble fields are non-renewable, as most plants only set seed prior to the onset of winter (Grime, Hodgson & Hunt 1989). In the absence of some form of cultivation exposing a previously concealed seed bank (Albrecht & Forster 1996), some depletion of seeds must occur. If winter food resources are limiting, either their initial density or that following depletion must determine the carrying capacity of the habitat (Goss-Custard & Durell 1990). Temporal variation across and within fields in linnet and reed bunting densities indicates that localized reduction of seed resources may limit the number of individuals sustainable in winter.

While strong correlative evidence of the relationship between food resources and a number of seed-eaters on winter stubble fields has been demonstrated, this relationship was not evident for all species studied. This may result from behavioural responses to predation risk, imperfect knowledge of winter diet for some species, lack of reliance on a seed-based diet and/or differential foraging abilities (Robinson 1997) For example, Robinson (1997) demonstrated that yellowhammers were significantly less efficient at finding buried grains than those on the soil surface. Thus, as the winter progresses and surface stocks are taken preferentially, buried grain is likely to make up a greater proportion of the remaining seed within a field and may not be as available as considered here. This may explain why, in this study, yellowhammer demonstrated significant correlation with grain resources at the start of the winter, but not by the end. Alternatively, yellowhammers may move to other habitats in late winter, as corn buntings do (Brickle 1998). As winter progressed, corn buntings moved from stubble fields to cattle-grazed grass and spring-sown barley fields, where cereal grain availability was greater due to feed provisioning for cattle and spring-sowing, respectively.

Conservation considerations

We have shown clear associations between food abundance on stubbles and densities of several seed-eating birds of conservation concern. Barley stubbles offered greater foraging opportunities than other stubbles for several species. However, other forms of stubble not included in this study may be of equal or greater conservation value to seed-eaters or invertebrate feeders (e.g. oilseed stubbles: Wilson, Tyler & Muirhead 1996; sugar beet: Donald et al. 2001). To maximize their value as winter foraging habitats for granivorous birds, stubbles are best provided on fields where:

  • (a) either immediate or long-term cropping history means that there is likely to be a substantial natural regeneration of a weed flora;
  • (b) previous cropping (e.g. barley) is likely to offer a sparse stubble with substantial areas of bare ground;
  • (c) prior herbicide use has been targeted to weeds of high agronomic concern, such as barren brome Anisantha sterilis L. and cleavers Galium aparine L., while sparing less damaging weeds of food value to granivorous birds.

Prior sowing of arable crops at greater drill widths (Odderskaer et al. 1997), and mechanical management to break up dense stubble covers by light cultivation of the stubble surface, may both be practical ways to improve foraging conditions on overwinter stubble fields; experiments could usefully test this hypothesis. Undersown stubble fields are likely to offer fewer foraging opportunities for granivorous species, but can be valuable habitats in the breeding season for invertebrate feeding species, such as grey partridge (Potts 1986), as many invertebrate prey species benefit from the lack of cultivation as the sown grass ley follows the arable crop in a mixed rotation (Barker, Vinson & Boatman 1997). Therefore it is important to consider the conservation of species needing opposing management strategies by developing management schemes at local or regional scales that benefit the whole bird community.

Supplementary material

The following material is available from http://www.blackwell-science.com/products/journals/suppmat/JPE/JPE730/JPE730sm.htm.

Appendix 1a. Correlations between all pairwise combinations of the predictor variables used in the multivariate analyses of frequency of occurrence on 122 fields for eight species in the weed models. Correlations were tested using either Spearman rank correlations (continuous variable vs. continuous variable), Mann–Whitney tests (two-level factor vs. continuous variable) or Kruskal–Wallis tests (> 2 level factor vs. continuous variable). Significant intercorrelations are in bold.

Appendix 1b. Correlations between all pairwise combinations of the predictor variables used in the multivariate analyses of frequency of occurrence on 27 fields for six species in the seed models. Correlations were tested using either Spearman rank correlations (continuous variable vs. continuous variable), Mann–Whitney tests (two-level factor vs. continuous variable) or Kruskal–Wallis tests (> 2 level factor vs. continuous variable). Significant intercorrelations are in bold.

Acknowledgements

We thank the 32 land-owners for access to their land, without which this research would not have been possible. The authors would also like to thank Allan Perkins, Phil Barnett and David Buckingham for their assistance with this research. Earlier drafts of the manuscript were improved by comments from Tony Morris, Paul Donald, Chris Elphick, Dan Chamberlain and one anonymous referee. This work was supported by the Biotechnology and Biological Sciences Research Council and RSPB.

Ancillary