1. Set-aside is arable land rested from normal intensive operations, usually providing, in summer, a relatively sparse, weedy or grass-dominated sward without pesticide or fertilizer inputs. Set-aside is therefore potentially attractive to breeding and foraging birds and is predicted to contribute to increased avian biodiversity on arable or mixed farmland. Set-aside mimics low intensity farmland within the heart of the industrial farm landscape, thereby allowing direct comparison with intensive crops regarding their respective values for the bird fauna.
2. In this study, bird abundance was compared between set-aside and nearby crops or grassland. A particular focus of the study was to identify the breadth or generality of any preferences across a suite of farmland species, using data from a broad representation of English farms. Thus, an extensive survey of birds utilizing fields, including set-aside, was conducted on 92 arable farms in England during 1996 and 1997. Each farm was visited four times in each summer, habitat details were recorded, and all birds seen or heard were mapped using a standard technique.
3. Field type preferences were examined across bird functional groups representing gamebirds, pigeons, crows, skylark Alauda arvensis, thrushes (Turdidae) and granivorous passerines (Passeridae, Fringillidae and Emberizidae). The relationship between bird abundance and field type was analysed using log-linear Poisson regression and compositional analysis.
4. Both analyses revealed that bird abundances were significantly higher on set-aside than on winter cereals for all six functional groups, and were highest on rotational set-aside for all functional groups except crows (which preferred grassland). Winter cereals or grassland were generally the least preferred habitat.
5. On farms where both rotational and non-rotational set-aside was present, preferences were strongest for rotational set-aside for all functional groups except crows (which preferred non-rotational set-aside). This underlines the differences between set-aside sward composition in influencing bird numbers.
6. The results show broad-scale preferences for set-aside over crops or grassland for species representing non-passerines, passerines, insectivores and granivores, for a wide representation of farms in England. For the majority of species, this preference implies that set-aside is utilized as a source of food, and the scale of this preference is impressive given that most set-aside was not managed specifically for bird conservation. However, not all types of set-aside were equally exploited by birds, as the strongest preferences were for natural regeneration rotational set-aside rather than the more structurally uniform non-rotational set-aside.
7. The results are important in the context of the potential loss of set-aside from the arable countryside, but also for the development of future agri-environmental schemes. We suggest that, to reverse population declines of many farmland birds in Britain, such schemes will need to be introduced on a wide geographical scale, like set-aside, but will also need to be carefully tailored, through advice to farmers, to maximize their potential to support bird species.
In recent years there has been growing concern about widespread loss of biodiversity in arable landscapes, in particular in the UK, mainland western Europe and the USA (Flade & Steiof 1990; Barr et al. 1993; Saris et al. 1994; Millenbah et al. 1996). In the UK a number of long-term monitoring programmes have shown substantial declines in the status of many groups of plants and animals (Firbank et al. 1991). For example, the results of the Countryside Survey 1990 (Barr et al. 1993) showed a marked reduction in hedgerow length and plant biodiversity in arable landscapes since 1978. Analysis of the British Trust for Ornithology's Common Birds Census, supported by two periodic atlas studies of bird distribution, also revealed that many of Britain's farmland bird populations have suffered serious long-term declines (Gibbons, Reid & Chapman 1993; Marchant & Gregory 1994; Fuller et al. 1995; Siriwardena et al. 1998), with declines less evident in other habitats such as woodland (Fuller et al. 1995). The species in decline represent a broad range of ecological needs and include birds like grey partridge Perdix perdix L., skylark Alauda arvensis L., song thrush Turdus philomelos Brehm and tree sparrow Passer montanus L. Similar declines in farmland birds are reported from elsewhere in Europe, for example, in Germany (Flade 1994) and the Netherlands (Saris et al. 1994).
Set-aside was introduced by the European Commission, as part of the Common Agricultural Policy (CAP), to reduce agricultural surpluses by removing areas of land from production. Until 1993, the uptake of voluntary set-aside included fewer than 2·5% of eligible farms in the UK (Evans et al. 1997). Following CAP reforms in 1992, subsidy payments to farmers (Arable Area Payments) effectively made set-aside widespread on arable land. In the summer of 1994, the set-aside rate was set at 15% of arable land for rotational set-aside and 18% of arable land for non-rotational set-aside (the latter a 5-year guaranteed income, encouraging farmers into the longer-term scheme). Set-aside then occupied approximately 0·70 million hectares of land. For 1995, the rates were reduced to 12% and 15%, respectively, and in 1996 a ‘flexible’ scheme allowed non-rotational set-aside to be returned to the crop rotation with a flat rate for all set-aside of 10%. This rate was reduced to 5% for the summer of 1997.
Management regulations aimed at preventing invasive ‘weeds’ from establishing required the destruction of the green cover on rotational set-aside by either cutting or cultivation, initially from 1 May in 1993, but then from mid-May in 1995 and eventually not before 1 July during 1997. Alternatively set-aside could be sprayed from mid-April with a non-residual herbicide. Up until early July, mechanized cutting of set-aside threatens the eggs and young of ground-nesting birds (Watson & Rae 1997; Poulsen, Sotherton & Aebischer 1998), whereas spraying kills the vegetation but may leave some cover intact (Wilson et al. 1995). Non-rotational set-aside was cut annually, by mid-August, with the cuttings left to degrade, so encouraging a thicker grass-dominated sward. In 1994, the rules allowed farmers to increase set-aside strips to 20-m rather than 10-m or whole fields, for greater flexibility and integration within the farm.
Although introduced as an agricultural measure, the presence of set-aside also provides areas of land that are relatively free from pesticides, fertilizer inputs and, in the case of most rotational set-aside, support over-winter stubble. The potential of set-aside, in terms of encouraging biodiversity on farmland, has been demonstrated by a range of studies. For example set-aside land supports more plant and invertebrate species than cropped land (Sears 1992; Wilson & Fuller 1992; Wilson et al. 1995) and provides a key foraging habitat for seed-eating passerines (Donald & Evans 1994; Evans & Smith 1994).
During the breeding season, studies of birds using set-aside have tended to concentrate on relatively few species, particularly skylarks and gamebirds, in small study areas (Manosa 1994; Poulsen 1996; Wilson et al. 1997; Poulsen, Sotherton & Aebischer 1998). Although these studies suggest that these species exhibit higher breeding densities and productivity on set-aside compared with nearby crops, the extent to which such findings can be generalized across species and across a broad geographical range remains unclear. The present paper is the first to report findings of an extensive study of the use of set-aside in summer by a wide range of bird species, including insectivores (e.g. Turdus species), seed-eaters (e.g. Carduelis species) and non-passerines (e.g. gamebirds and Columba and Streptopelia species), on more than 90 farms in lowland England, UK. On the basis of this study, the application of set-aside and related non-cropped, agri-environmental schemes can be tailored better to maximize their potential to support bird populations on farmland.
The ornithological aims of this study were to: (i) assess the relative use of set-aside by birds compared with nearby crops or grassland; (ii) determine which groups of bird species were most closely associated with set-aside on lowland arable farms; and (iii) compare the use made by birds of two distinct types of set-aside, rotational set-aside (where the fallow vegetation is in its first year) and non-rotational (which remains in place for more than 1 year and where the fallow vegetation is at least in its second summer).
The study sites comprised a random sample of 200 farms in England, stratified by set-aside type (rotational/non-rotational) and geographical area (arable and mixed farming regions). Volunteer observers surveyed 92 of these sites in 1996 (40 and 43 rotational and non-rotational set-aside sites, respectively, nine with both set-aside types) and 63 sites in 1997 (29 and 27 rotational and non-rotational set-aside sites, respectively, seven with both set-aside types; Fig. 1). The 1997 sample was a subsample of the farms surveyed in 1996. On each farm, an observer was asked to define a survey plot by locating a pre-selected field of set-aside from supplied farm crop maps, plus 5–10 surrounding fields (which may also have contained set-aside). The mean plot area was 44 ha (range 15–107 ha), within which the proportion of set-aside varied between 7% and 75% (mean 22%). All survey plots contained less than 10% woodland within their boundary. Some set-aside land was used, within the rules, to produce raw materials for non-food purposes (such as oil from ‘industrial oilseed rape’). These crops were not included in the survey, as their management (e.g. pesticide and fertilizer applications) was biologically indistinct from similar food crops.
Survey bird data
Observers made four visits to a farm plot between April and July (one visit per month with at least 2 weeks between visits) in both 1996 and 1997. Each farm was surveyed by a different observer. On each visit to the farm site, the observer walked around all field boundaries, recording the field or boundary location of all birds seen or heard onto 1 : 2500 maps of the plot. This method was used to achieve a balance between wide farm coverage (through easy access to fields via their boundaries) and adequate precision in bird detectability. The survey was operated in conjunction with an intensive field-transect study, on fewer farms, reported elsewhere (Henderson, Vickery & Fuller 2000). This study showed that detectability was not a particular problem for large species such as crows and pigeons, or singing skylarks, or for species that concentrate their foraging efforts towards the field margins, such as thrushes, buntings and chaffinch Fringilla coelebs L. (Henderson, Vickery & Fuller 2000). However, it is acknowledged that some smaller or cryptic species such as Cardueline finches or partridge species may be less detectable in larger fields, and this matter is given some attention in the analysis and its interpretation below. This counting technique is a modification of the standard territory mapping approach used in the British Trust for Ornithology (BTO) Common Bird Census, but is used here to give relative, not absolute, estimates of numbers of birds on different fields and boundaries. The field methodology uses strict codes and guidelines to record bird activities (such as singing, displaying, feeding and flying) that require observers to distinguish between independent records and probable duplicates (i.e. movements of individual birds), which may indicate whether a bird is using the habitat in question for feeding or breeding purposes (Marchant et al. 1990). Each observer was required to finish a visit within 3 h and before 10.00 hours British Summer Time. Observers were requested to vary the survey route between visits to ensure that diurnal variation in bird activity did not bias the recorded distribution. No visits were made in heavy or persistent rain or in wind greater than force 4. In the current paper each individual bird record is termed a ‘count’ and only counts that observers could definitely attribute to independent birds were used in the analyses. A bird that moved between two fields or boundaries during a visit was scored 0·5 records for each field or boundary used. Birds in flight, including hirundines and swifts Apus apus L., which did not take-off from or land in a particular field or boundary, were excluded from the analyses, except for hunting birds of prey (e.g. kestrel Falco tinnunculus L.), which were retained within the data set. At each site and visit, the sum of the counts of each species on each field was used as the basic unit to compare relative bird densities and distributions between field types.
Bird species were assigned to functional groups because many individual species were recorded in low numbers. Sample sizes, even with the large numbers of farms involved, were too small to permit statistical analysis of most individual species. The grouping retained the foraging and ecological variation present in the original array of species by allocating species to functional groups on the basis of taxonomic and ecological factors, with emphasis on separating predominantly granivorous species from insectivorous species. Functional groups were as follows: gamebirds; pigeons; crows (Corvus species only); thrushes (Turdus species, i.e. ground-feeding insectivorous passerines); and granivores (ground-feeding seed-eating passerines). One species, skylark, was sufficiently numerous to be analysed separately. Appendix 1 lists all analysed species and their numerical contributions to their allocated functional group.
Survey habitat data
Several papers have emphasized the importance of the physical characteristics of hedgerows in determining bird abundance, in particular hedge height, tree frequency and the presence of vegetated hedge fringes (i.e. ‘grassy strips’) (Green, Osborne & Sears 1994; Parish, Lakhani & Sparks 1994, 1995; Macdonald & Johnson 1995). Observers therefore estimated the average height of each boundary section (a square field would have four boundary sections) and the width of the strip of vegetation lying between a boundary and the crop (‘boundary strip width’, including grassland where marginal strips were visible on ‘cropped’ silage and hay fields) and counted the number of trees (≥ 5 m tall) in each boundary section (‘tree frequency’). Boundary lengths (m) and field areas (ha) were extracted from the plot maps. Farmers provided crop details, which were classified into six categories, as follows. (i) Rotational set-aside: dominated by natural regeneration of cereal volunteers (from the previous crop) and indigenous flora (i.e. 95·5% of fields) or, rarely, a sown grass cover (i.e. 4·5% of fields). (ii) Non-rotational set-aside: natural regeneration or a sown grass cover, at least 2 years old. (iii) Winter cereal: autumn-sown wheat, barley or oats. (iv) Spring cereal: spring-sown wheat, barley or oats. (v) Non-cereal: brassicas (including oilseed rape), legumes, root crops, linseed, maize or other non-cereals. (vi) Grassland: combined ley grass (whether grazed or non-grazed), permanent pasture (whether grazed or non-grazed), silage or hay. To help interpret levels of bird detectability between field types, observers estimated average crop height by categories for each field on each visit (1 = >10, 2 = 10–25, 3 = 25–50, 4 = 50–100, 5 = >100 cm). This analysis was used to identify extreme differences in crop height that could have confounded the bird count data.
The relationship between bird abundance and field type was analysed using two techniques: a log-linear analysis (Brown & Rothery 1993) and a compositional analysis (Aebischer, Robertson & Kenward 1993; Elston, Illius & Gordon 1996). The log-linear models were appropriate for data that were Poisson distributed, i.e. for non-negative data where the variance increased and was equal to the mean. Log-linear models allowed independent analysis of both 1996 and 1997 data and gave the multiplicative effects of field type on bird densities relative to winter cereals. Compositional analysis was not reliant on model fits for the interpretation of the results and was used to analyse how birds were distributed across field types in proportion to each field type's availability. Sites with rotational and non-rotational set-aside were analysed in separate models but sample size restricted the analysis to the larger 1996 data set. Both methods controlled for farm effects on bird numbers that might have arisen through differences between observers, geographical location or the farm management regime.
The relationship between bird density and field area was also analysed by Spearman rank correlation, to provide a measure of the potential bias in detectability related to field size for each functional group. In addition, differences between field types in the average estimated vegetation height were analysed using Kruskal–Wallis non-parametric tests.
To compare relative bird abundances on the six field-type categories defined previously, generalized linear models with a log-link function and Poisson error term were fitted to the functional group summed count data from each individual field on each farm site, for each of the four visits, using a repeated-measures analysis for each visit (Genmod procedure; SAS Institute Inc. 1996). Models included controls for boundary explanatory variables, i.e. hedge height, boundary tree frequency and the boundary strip width. The model fitted to each functional group count was as follows:
N = exp(I + ci + sj + (h × height) + (t × tree freq.) + (m × width) + a) + ε
where N = number of individuals in each functional group on each field type; I = intercept term; ci = factor representing field type i (where i = one of six field types defined above); sj = factor representing farm site j (j = 1 − 92 farms), used to assess between-farm differences in bird abundance; h = coefficient of boundary height explanatory variable (to assess the effects of boundary height, measured per field); t = coefficient of boundary tree frequency explanatory variable (to assess the effects of trees in boundaries, measured per field); m = coefficient of boundary strip width (margin) explanatory variable (the effect of the boundary strip of vegetation, non-cropped, measured per field); a = log (area) offset variable for each field type (to control for field area); ε = Poisson error term.
All models were fitted relative to winter cereal, with parameter significance tests calculated with respect to this crop. Winter cereal was selected because it occurred on more farms than any other crop. The square root of the scaled deviance/degrees of freedom was used as an over-dispersion factor in parameter significance tests (SAS Institute Inc. 1996). A repeat analysis on a subset of field types was carried out using April and May counts only. During this period there was least contrast between crops and set-aside in terms of vegetation height (analysed below), with birds also being visible along crop rows. A third analysis compared bird counts on rotational and non-rotational set-aside relative to winter cereals on 10 farms where both set-aside types were present. These data were again analysed using the Poisson regression procedure above. All models returned type 3 likelihood-ratio (LR) significance values for each factor in turn, while controlling for all other factors in the model.
Analysis of whether or not birds were distributed across field types in proportion to the availability of each field type on a farm, was subject to the unit-sum constraint, whereby proportions at each site sum to 1 (Aebischer, Robertson & Kenward 1993). In view of this, proportional data were analysed using a compositional analysis, which breaks this constraint through a log-ratio transformation (Elston, Illius & Gordon 1996). This procedure allowed estimation of the relative preferences of birds for field types and the statistical significance between these preferences. Following Aebischer, Robertson & Kenward (1993), the data were first transformed thus:
where xi= component proportion (i.e. the proportion of field type I or birds on I); xj= component field type proportion used as denominator; f = number of available field types.
The resultant transformed vector was linearly independent of the denominator field type, xj. In accordance with Aebischer, Robertson & Kenward (1993), zero counts of birds were replaced with the small value 0·001 before log-ratio transformation. Given f field types, an f by f matrix was constructed from each vector of proportional data, as shown below.
Each study site had a matched pair of proportional vectors, i.e. the proportion of birds on field types and field types on the farm. Therefore, paired matrices were generated for each functional group on each farm, one containing data on field use by birds (Fu) and the other on field availability (Fa). Subtracting Fu from Fa yielded a matrix of log-ratio differences, d, each element of which was equivalent to zero under the null hypothesis. This matrix was averaged across the farms containing the set of field types being compared. Significant departure from random for each element of this mean matrix was evaluated using randomization tests involving 999 permutations of the data. Values significantly (α = 0·05) greater or less than zero identified field type preference or avoidance relative to the denominator type. Field types were then ranked in order of use by functional groups, following Aebischer, Robertson & Kenward (1993).
The occurrence of spring cereals was too infrequent for inclusion in the compositional analysis. Also, the incidence of rotational and non-rotational set-aside on farms was largely mutually exclusive. Therefore, two sets of comparisons were made using winter cereal, non-cereal and grassland plus either rotational or non-rotational set-aside, for each of six functional groups (gamebirds, pigeons, skylarks, thrushes, crows and granivores).
All log-linear models fitted to functional group count data had a degree of over- or under-dispersion, as indicated by the deviance/d.f. values being greater or less than 1 (Table 1). Compensating for this with the dispersion factor, all models for functional groups, in both years, had highly significant field-type effects (P < 0·0001, when all other explanatory variables were controlled for; Genmod Type 3; SAS Institute Inc. 1996). Other significant explanatory variables (P < 0·05) in both years included farm type (all functional groups); a positive effect of hedge height for all functional groups except gamebirds, pigeons and granivores in 1996 (no effect) and skylark (negative effect); tree density within boundaries for crows, pigeons, thrushes (positive effect) and skylarks (negative effect); and a positive effect of boundary strip width on pigeons and skylarks.
Table 1. Log-linear model comparisons of bird abundance (in functional groups) on grassland (G), non-cereals (NC), non-rotational set-aside (NR), rotational set-aside (R) and spring cereal (SC), relative to winter cereal (WC; where the relative abundance estimate = 1). (a) Covers the period April–July; (b) the period April–May. Significance tests were calculated relative to winter cereal (*P < 0·05, **P < 0·01, ***P < 0·001). The number of fields used was 310 in 1996 and 212 in 1997. Field types with the highest and lowest parameter estimates are summarized in the final two columns. Model dispersion factors are given by deviance/d.f. (dev./d.f.)
Exponential of parameter estimates
These groups were analysed using a binomial function and logit error term to handle a high proportion of zero counts or clumped data. Here the exponential values represent the probability of occurrence on each field type (where winter cereals = 0·50).
The differences in bird count data between field types were consistent across functional groups for each field type relative to winter cereals (Table 1a). These data revealed that relative bird abundance (i.e. count data with field area controls) on rotational set-aside was higher than on any other crop type for all functional groups and years except crows in 1996 (preferring grassland; Table 1a). Bird abundance on non-rotational set-aside was significantly higher than on winter cereals in all 12 cases although, after rotational set-aside, non-rotational set-aside was ranked highest in only eight of the comparisons. Winter cereals supported the lowest bird densities in at least 1 of the 2 years for each functional group and in 10 out of 12 comparisons. The number of farms with spring cereal was low (23 in 1996 and 20 in 1997), which may have contributed to model inconsistencies such as this spring cereal receiving the second highest and second lowest abundance estimates in different years for thrushes. Table 1b presents similar results for April and May counts only, when there was minimal contrast between the vegetation height of set-aside and cereals (Table 2; although the vegetation height on grassland was lowest among field types throughout the survey period). The results in Table 1b display the major field types only, rotational set-aside, non-rotational set-aside and grassland with respect to winter cereals, showing that the pattern of preferential selection for set-aside, over-winter cereals in particular, was retained for all functional groups. In eight out of 12 cases, winter cereals returned the lowest estimate of relative abundance. Rotational set-aside produced the highest abundance estimate in 10 out of 12 models. Grassland was the preferred field type for crows in 1996 and thrushes in 1997.
Table 2. The mean vegetation height category for four field types in each sample month April–July, for 1996 and 1997 combined (Bonferroni corrected α = 0·006; n= number of fields)
Mean height category (in sample month)
1. Rotational set-aside
2. Non-rotational set-aside
3. Winter cereal
P < (Kruskal–Wallis)
P < (excluding grassland; Kruskal–Wallis)
There were no negative correlations (significant or otherwise) between bird density and field size among functional groups in any field type. There were significant positive correlations for all functional groups (P < 0·05), except gamebirds and thrushes, on grassland, and for skylarks on all field types (P < 0·001), including winter cereals (Spearman's rho = 0·23, n = 287, P < 0·0001). The density of pigeons increased with field size in non-cereals (Spearman's rho = 0·22, n = 101, P < 0·03). These results do not support the notion that lower detectability in larger fields was a systematic source of bias during the analysis of field-type preferences above.
On the 10 farms where rotational and non-rotational set-aside were both present, there were significant differences in the relative abundance of birds between field types for all functional groups [gamebirds likelihood ratio (LR): χ22 = 112·6, P < 0·001; pigeons LR: χ22 = 50·3, P < 0·001; skylark LR: χ22 = 77·6, P < 0·001; thrushes LR: χ22 = 36·2, P < 0·002; crows LR: χ22 = 57·4, P < 0·001; granivores LR: χ22 = 94·6, P < 0·001; scale deviance/d.f. range between 0·6 (thrushes) and 6·7 (crows)]. Rotational set-aside supported the highest relative abundance of birds for all functional groups except crows (for which non-rotational set-aside was marginally the highest), with winter cereals supporting the lowest abundances in all cases.
In the compositional analysis, there was a significant difference in usage between the first and last ranked field types for each functional group (at α = 0·05: Table 3). For five of the six functional groups, rotational set-aside was ranked highest, and it was ranked second highest by the sixth functional group. Non-rotational set-aside was ranked first for four of six functional groups and ranked second for the remaining two. The least preferred field type was either grassland or winter cereal, both of which were used significantly less than set-aside by all groups throughout the analysis. Among the functional groups, gamebirds were the main exception to the general trend, favouring non-cereals to both rotational and non-rotational set-aside, although the difference between field types was not significant (P > 0·05). The preference by crows for grassland (with non-rotational set-aside ranked second) was consistent with the log-linear analysis above.
Table 3. Results from the compositional analysis based on 1996 data. Field types are ranked in order of preference (1, highest; 4, lowest) for six functional groups, with significant differences between field types to the right marked * (α = 0·05, from randomization tests). The difference between ranks 1 and 4 is significant in each test (i.e. P < 0·05). The table is in two parts, (a) comparing rotational set-aside (upper) and (b) non-rotational set-aside (lower) with crops or grassland
Field type preference rankings
(a) Farms with rotational set-aside (n = 40 farms)
(b) Farms with non-rotational set-aside (n = 43 farms)
Winter cereal *
The results of this study show that set-aside in general, and rotational set-aside in particular, was used in preference to cultivated crops or grassland throughout the summer by a broad range of bird species found on English arable farmland. Several detailed autecological studies have emphasized that characteristics of set-aside, such as winter stubbles and weed-rich fields, may meet key ecological needs of species such as the skylark and gamebirds, and have demonstrated preferences by these birds for set-aside (Potts 1986; Donald & Evans 1994; Evans & Smith 1994; Wilson et al. 1997; Chamberlain et al. 1999b). However, to date no studies have demonstrated such broad generality of preference for set-aside in summer by non-passerines, passerines, insectivores and seed-eating bird species, across such a wide geographical area within the UK.
These preferences for set-aside were demonstrated using two different analytical approaches. Log-linear analysis resulted in models indicating that, in each of two years (1996 and 1997), bird densities on rotational set-aside consistently exceeded those recorded on neighbouring fields, particularly winter cereals. Compositional analysis, which was not reliant on model fits for the interpretation, showed that, in terms of proportional occupancy by birds across farms, rotational set-aside was usually ranked highest amongst other field types. This was also true for non-rotational set-aside, although to a lesser extent.
Although our study design and analysis accounted for the possible effects on bird densities of sample time, geographical location, bird breeding biology and differences between observers, there are additional factors that could confound the findings. They are, first, the potential non-random placement of set-aside (particularly non-rotational) within farms, and, secondly, differences in the detectability of birds on set-aside compared with crops. These potential biases are considered below.
It has been suggested that non-rotational set-aside may often be located on agronomically and/or spatially marginal areas of a farm (Crabb, Parham & Dauven 1998). Such areas may offer different features than cultivated areas, for example they may receive lower pesticide inputs or may be less well drained. This could make areas of non-rotational set-aside (and grassland too) relatively attractive to birds. However, this seems an unlikely explanation for the preferences shown for rotational set-aside, which changes location from year to year. In addition, the selection of non-rotational set-aside over grassland by all functional groups except crows (which are known to prefer to forage in grassland; Goodwin 1986) is probably genuine as grassland is also likely to occupy agronomically marginal land within largely arable systems (O’Connor & Shrubb 1986).
Differences in the detectability of birds, between set-aside and other field types, could result in a systematic underestimation of birds using certain crops or grassland. From a concurrent field-transect study of set-aside and adjacent land, we know that the main species or groups of species likely to be affected by such biases are the smaller or inconspicuous species that forage away from boundaries, such as gamebirds (e.g. grey partridge), skylark and Cardueline finches (especially linnet) (Henderson, Vickery & Fuller 2000). Among the other groups in the present study, crows and pigeons (pigeons were numerically dominated by wood pigeons Columba palumbus L.) are relatively conspicuous on farmland, while thrushes and gamebirds have a very strong field marginal bias to their foraging distribution (Henderson, Vickery & Fuller 2000). Skylarks, meanwhile, tended to be detected by their song (i.e. territorial males) while flying above the vegetation in all field types. We feel that biases in the detectability of these latter groups and skylark were minimal using the techniques employed by the present study. A strong field-edge foraging distribution is also recognized for house sparrow Passer domesticus L., tree sparrow, chaffinch and buntings (Emberizidae) and to a lesser extent greenfinch Carduelis chloris L. and goldfinch C. carduelis L. (Henderson, Vickery & Fuller 2000). Thus, among the granivorous species, linnets, which formed 35% of the counts within the group, were the main source of potential bias in detectability.
The strong preference for set-aside (and rejection of winter cereals) by all functional groups during early summer (April–May), was further evidence that detectability was not prominent in defining differences in abundance between field types. At this time there was very little contrast between crops and set-aside, in terms of vegetation height among the fields surveyed (Table 2). Visibility on grassland should have been especially favourable. Detectability in young crops is aided by good visibility along crop rows compared with the scattered vegetation growing on set-aside. Only among non-cereals, and oil-seed rape in particular, was the crop likely to be especially dense in early summer. Moreover, evidence for strong biases in detectability between crops was expected to be reflected in negative relationships between bird density and field area within field types. However, no such correlation (significant or otherwise) was found for any functional group, and indeed all significant correlations between bird density and field area were positive (i.e. birds tended to ‘favour’ larger fields).
Overall, it is unlikely that detectability introduced any systematic bias to the results among the functional groups observed. Thus, the selection for set-aside, especially rotational set-aside, across functional groups is considered to reflect a genuine preference for this field type over crops and grassland. With the exception of skylarks, which nest in open fields, birds utilizing fields were likely to be foraging rather than nesting there. Several studies provide evidence to suggest that set-aside may provide food resources for farmland birds that have become increasingly scarce during the last 30 years (Fuller et al. 1995). For example, set-aside has been shown to support a greater abundance of invertebrates (Kennedy 1992; Moreby & Aebischer 1992; Poulsen, Sotherton & Aebischer 1998) and weed seeds (Draycott et al. 1997) than neighbouring crops. Indigenous plants, such as Stellaria media, Senecio vulgaris, Matricaria spp. and Taraxacum officinale, establish quickly on first-year set-aside (Clarke & Cooper 1992; Wilson & Aebischer 1995; Hansson & Fogelfors 1998) and provide food for omnivorous and seed-eating birds (Wilson, Arroyo & Clark 1996) and habitats for a range of invertebrates (Rands 1985, 1986; Potts 1986, 1991). In addition there is evidence that foraging efficiency of some bird species is higher in set-aside than within other crops (Manosa 1994; Poulsen 1996). Thus it seems highly likely that the preference shown for set-aside in the present study is attributable to the fact that this habitat provides increased foraging opportunities for a broad range of bird species.
Although bird density was higher on set-aside in general than on winter cereals, the preferences shown by most functional groups were stronger for rotational than non-rotational set-aside. This preference may be related to differences in sward structure and composition, as there was no clear geographical distinction between rotational and non-rotational set-aside sites (Fig. 1). In the present study, rotational set-aside had a much more patchy, species rich and structurally complex sward than non-rotational set-aside (Critchley & Fowbert 1998; Fowbert & Critchley 1998). Unless managed to maintain early successional stages (Hansson & Fogelfors 1998), set-aside develops a dense perennial-dominated sward (Wilson 1992; Fisher, Davies & Christel 1994) and vegetation density is known to be a key factor influencing foraging efficiency and habitat preferences in some ground-feeding birds (Henderson & Evans 2000). For example, sparse grassland of 1–2 years old has been shown to support a greater diversity and relative abundance of bird species than older stages of growth, a difference attributed to the ease of foraging in more open, younger, swards (Millenbah et al. 1996). Similarly, Watson & Rae (1997) suggested that the high usage of first-year set-aside by grey partridges, skylarks and some wader species was related to the varied tall but sparse vegetation structure, offering increased feeding opportunities as well as cover and ground access. These differences suggest that the benefits of set-aside for birds are not simply derived from the removal of land from production but depend crucially on the nature of the sward that develops.
Set-aside effectively became compulsory in Britain in 1992. It has been shown to be a preferred foraging habitat by a range of species both in winter (Donald & Evans 1994; Evans & Smith 1994; Poulsen 1996; Wilson, Taylor & Muirhead 1996; Buckingham et al. 1999) and summer (this study). For some species, notably the skylark, breeding success is also higher on set aside than on intensively managed cereals (Wilson et al. 1997). However, although individual studies have shown high densities and productivity of birds on set-aside, whether this simply reflects a change in distribution of birds within the arable landscape (see below) or whether increases in breeding productivity are sufficient to make a difference at the population level, remains unknown.
A key question is whether the introduction of set-aside has actually resulted in a slowing down or reversal of population declines of farmland birds. Long-term monitoring data from the BTO's Common Birds Census (CBC) provides one way of addressing this question. We considered the skylark to be the species most likely to have shown a positive response to set-aside at the population level. The species is strongly territorial, easily censused and dependent almost entirely on field quality, rather than boundary quality, for feeding and nesting. It is also one of the species for which we have the best evidence that farm management can affect breeding success (Wilson et al. 1997). However, a continuing shallow decline in the national trends in the UK for skylark (Siriwardena et al. 1998) may indicate a redistribution of territories away from crops onto set-aside, without an overall increase in territory density. Nationally, skylarks have declined dramatically since the 1970s, although the rate of decline has slowed during the late 1980s and through the 1990s. This slower rate of decline began before voluntary set-aside was introduced in 1988 and well before set-aside was commonplace from 1992 onwards, and is probably best accounted for by the parallel slowing up in the rate of change in agricultural intensification (Chamberlain et al. 1999a) rather than by any active intervention. It is certainly possible that the introduction of set-aside has helped to ease the rate of decline of the skylark. However, the absence of any clear upward trend in numbers has important implications for conservation of this and other farmland bird species.
The most likely explanation for the lack of any clear ‘population effect’ of set-aside is that there simply has not been a sufficient area of set-aside in place for long enough to result in a population increase large enough to be detected by the CBC. At its peak, between 1993 and 1995, set-aside nominally occupied between 15% and 18% of arable farmland, reduced to 5% in 1997, probably representing one of the single largest changes in farming practice witnessed over 2 years (Evans 1997b). In addition, as much as 50% of set-aside was managed either as annually mown non-rotational set-aside or as industrial set-aside (Evans 1997b), both of which were largely unsuitable as breeding habitat for skylarks. The actual area of set-aside with the potential to influence population levels may therefore have been as little as 8% at its peak.
These results suggests that if agri-environment measures are to have any impact on declining farmland birds at the population level they will need to be not only introduced on a very wide geographical scale but also to be managed to maximize their value to birds or wildlife in general. A new agri-environment scheme, the Ministry of Agriculture, Fisheries and Food's Arable Stewardship Scheme, is currently being piloted in two areas in the UK. Unlike set-aside, this scheme has been designed with specific environmental objectives, so the potential benefits for biodiversity are greater than through set-aside. However, the extent to which these benefits are realized will depend crucially on the uptake. The success with which the prescriptions within the scheme deliver environmental objectives requires careful monitoring of the flora and fauna they are designed to benefit, the patterns of uptake by farmers, and the reasons for these patterns.
The study would not have been possible without the participation of the volunteer observers. We are extremely grateful to the observers, to the BTO regional representatives and to the farmers who gave permission for the survey to be conducted on their land. We have greatly benefited from discussions with Les Firbank, Peter Rothery and Mark Hill of the Institute of Terrestrial Ecology (ITE), Nigel Critchley, Chris Britt and James Clarke (ADAS) and Tony Hughes (MAFF). For their help at various stages of the project we also thank Graham Austin (computing), Julianne Evans (pilot survey), Steve Freeman (statistical advice), Nigel Clark, Nick Carter, Sophie Foulger and Nicola Read of the BTO, Noranne Ellis and Ruth Swetnam of ITE, and John Fowbert and Naomi Jones of ADAS. This research programme was funded by the Ministry of Agriculture, Fisheries and Food (MAFF).
Received 5 February 1999; revision received 17 December 1999
Table 4. A list of species and allocated functional groups used in the analyses above. The total number of sightings on all fields (across all four farm visits) in 1996 and 1997 for each species are also presented.
Gamebirds: all gamebird species, mostly with artificially maintained populations.
Pigeons: all non-feral species of Columba and Streptopelia genera.
Crows: Corvus species, rook, jackdaw and carrion crow.
Thrushes: blackbird, song thrush and mistle thrush; boundary-based insectivores which regularly feed on the ground away from boundaries.
Granivores: boundary-based seed-eating species that regularly feed on the ground away from boundaries; comprising Emberizidae, Fringillidae and Passeridae.