1Over the last 25 years, populations of seed-eating birds have declined severely over most of western Europe. Local extinctions have occurred in grassland-dominated areas in western Britain, which may be influenced by loss in habitat diversity and a decline in the amount of arable cultivation.
2We used the large-scale British Breeding Bird Survey of 1998 to investigate the importance of arable habitat within grassland landscapes for 11 common seed-eating birds and four similar sized insectivores. Generalized linear models were used to model the number of birds recorded in agricultural habitat within survey squares as a function of the amount of arable habitat present.
3Numbers of grey partridge Perdix perdix, skylark Alauda arvensis, tree sparrow Passer montanus, corn Miliaria calandra and reed buntings Emberiza schoeniclus, yellowhammer Emberiza citrinella and whitethroat Sylvia communis increased with the amount of arable habitat present in a survey square; the numbers of house sparrow Passer domesticus, four finch species, dunnock Prunella modularis, robin Erithacus rubecula and blackcap Sylvia atricapilla did not.
4The positive association between numbers of some species and arable habitat within 1-km squares was strongest where arable habitat was rare in the surrounding area, and weakest or even reversed when arable habitat was common. These results demonstrate the scale-dependence of bird–habitat associations in agricultural landscapes, only demonstrable where data are available at fine grain over large geographical areas.
5These results support the hypothesis that range contractions (i.e. local extinctions) of some granivorous species have occurred because of contraction in arable cultivation. The loss of arable habitat where it is scarce may be causing declines in some areas, even though intensification of arable management is thought to be the main cause of declines elsewhere. Agri-environment schemes may need to vary between regions, for example to encourage arable cultivation in pastoral areas.
At the turn of the century, British farmland was mostly of a mixed character, with crop and animal husbandry coexisting on most farms. Much agricultural land fell into disuse during the economic depression of the 1930s, but with the increase in agricultural activity in the late 1940s the farming landscape became increasingly polarized, with arable predominating in the east of Britain and pasture-based farming in the west (Stoate 1995). This was largely caused by an increase in mechanization, which removed the need to grow forage crops for horses on arable farms, whilst an increase in use of agricultural chemicals removed the need for non-cereal crops and fallows, which maintained soil fertility and prevented pest build-up (Stoate 1995, 1996). These trends are continuing as the necessity for capital-intensive machinery encourages specialization in either crop or animal husbandry. Consequently, arable habitat has become increasingly scarce in pastoral landscapes, with the loss tending to be greatest in the most grassland-dominated areas (Table 1).
Table 1. Loss of arable land in grass-dominated areas (arable : grass ratio < 0·5) between 1970 and 1997 (data from MAFF statistics). The data for Scotland measure change 1978–97 and no account has been taken of slight changes in recording methods during this period
% arable 1970
% arable 1997
Upland Northern Ireland
Lowland Northern Ireland
The declines in farmland bird populations, particularly granivorous passerines, are likely to have been caused by changes in arable management because arable farmland holds significant proportions of their populations (Gregory & Baillie 1998; Fuller 2000). Changes in the timing of cereal sowing, reducing the availability of seed-rich winter stubbles (Wilson, Taylor & Muirhead 1996), and reductions in food supply caused by increased herbicide use (Campbell et al. 1997), are likely to have been particularly important, although most of the changes are likely to be interrelated (Chamberlain et al. 2000). Although populations of many of these species have declined more rapidly in arable than pastoral areas, the extent of local extinctions in western Britain has been much greater than in the arable east of the country, because bird densities were initially lower in grassland areas (Chamberlain & Fuller 2000).
Vickery et al. (1999) collated data on national bird abundance to assess differences between grassland, mixed and arable habitats. They found that the majority of species occurred most abundantly in areas of mixed farming, particularly in winter. Detailed autecological studies have shown that the breeding densities of some species, notably skylark Alauda arvensis (Jenny 1990) and corn bunting (Ward & Aebischer 1994), are directly related to the diversity of crops available at a farm scale. Thus, the loss of arable habitat in grassland regions may have contributed to the declines of farmland birds in these areas and, conversely, support for arable habitat in these areas could help the conservation status of these species.
In this study we aimed to test whether increasing arable cultivation in pastoral landscapes influences the local abundance of 11 species of granivorous bird that are particularly dependent on arable habitat (skylark, chaffinch Fringilla coelebs, greenfinch Carduelis chloris, goldfinch Carduelis carduelis, linnet Carduelis cannabina, tree sparrow, house sparrow Passer domesticus, yellowhammer Emberiza citrinella, reed bunting Emberiza schoeniclus, corn bunting and grey partridge). Populations of all but chaffinch, greenfinch and goldfinch have declined severely (> 35%) since 1970 (Siriwardena et al. 1998). We also included four insectivores of similar body mass that are common on farmland: two residents, dunnock Prunella modularis and robin Erithacus rubecula, and two migrants, whitethroat Sylvia communis and blackcap Sylvia atricapilla. These were used as reference species to indicate whether arable cultivation provides additional resources specifically for the seed-eaters or some other, more general, benefits. Of these insectivores, only the dunnock has shown a slow population decline; the other three have increased in numbers since 1970. Nomenclature follows BOU (1992).
The analyses of bird numbers used data from the 1998 British Trust for Ornithology/Joint Nature Conservation Committee/Royal Society for the Protection of Birds (BTO/JNCC/RSPB) Breeding Bird Survey (BBS). This is a volunteer-based survey organized by the BTO and is designed to provide an index of annual changes in bird populations across the whole of the UK (Noble et al. 1999). Sample plots (1 × 1-km squares) are selected randomly, with stratification according to available observer effort, from within 83 sampling regions. In 1998, 2173 sample squares were surveyed. BBS data from 1997 were also analysed to provide a contrast with 1998 because numbers of breeding birds were relatively low following a poor breeding season in 1996 (Crick et al. 1998). The 1997 data yielded very similar results to 1998, so are not presented here.
Each year, BBS squares are visited three times. An initial visit is made in early April to record habitats according to a standard set of categories, with a hierarchical set of subcategories to provide finer definitions (Table 2). Two morning visits (beginning at 06·00–07·00 and lasting around 90 min to coincide with peak bird activity) are made during the breeding season (one in April/early May and one in late May/June) to count birds present along two parallel transect routes traversing the square. Each transect is approximately 1 km long and is divided into five equal recording sections; habitat and bird numbers are recorded separately for each of the 10 transect sections in the square. The distance of each bird sighting from the transect line is recorded in one of three bands (0–25 m, 25–100 m and > 100 m). In this study, only birds in the first two distance bands (within 100 m of the transect line) were considered. Juvenile and immature birds were excluded, thus counts refer to potentially breeding adults only. Birds flying over the square are noted separately, but, with the exception of display-flighting skylarks, which are recorded in the appropriate distance band, were excluded in this study. The number of individuals of each species occurring in the square was estimated as the maximum of the two counts.
Table 2. Habitat coding system used; level 2 codes for non-farmland habitat are not given. From Crick (1992). Additional levels allow for finer-scale habitat details on, for example, crop type and hedgerow presence
Mixed grass/tilled land
For the purposes of this study, birds occurring on non-farmland transects (i.e. not code E; Table 2) were excluded to reduce variation in bird numbers caused by the presence of other habitats. Squares with < 5 transect sections coded as farmland (E) were also excluded, leaving a total of 1350 BBS squares. Throughout, the term ‘bird numbers’ refers to the number of birds seen within 100 m of the transect line in agricultural transect sections for a given (agricultural) BBS square. The amount of arable (= annually cultivated) habitat in a square was estimated as the number of transect sections containing tilled land (code E4 in the BBS), mostly cereals and root crops. Transect sections where the habitat was recorded as mixed farming (E3) were assumed to contain, on average, 50% arable habitat and were thus counted as half an arable transect section. No correction was made for differences in bird detectability between habitats, which were likely to be small because of the open nature of the habitats considered; strictly, the counts provide a relative index of abundance only. In recording crop type, not all surveyors differentiated between winter- and spring-sown cereals. In analyses involving these variables, we included only squares in which these were explicitly differentiated or where no arable was recorded (eliminating squares where these habitats were present but not recorded).
To control for differences in landscape type, Britain was divided, on a county basis, into three regions, pastoral, mixed and arable, using the annual UK Ministry of Agriculture, Fisheries and Food (MAFF) June Census statistics (MAFF 1998) (Fig. 1). Pastoral counties (32) were defined as those with an arable : grass ratio of < 0·5 (i.e. containing more than twice as much grass as arable). The nine counties with arable : grass ratios > 2·5, which are dominated by intensive arable cultivation, formed the arable group, and the remaining 24 counties were classified as mixed farmland counties. The three regions held 517 (pastoral), 329 (arable) and 512 (mixed) BBS squares. The 32 pastoral counties were further split into two groups: 14 upland counties (161 squares) with arable : grass ratios of < 0·1, which consisted mostly of upland grass pasture and rough grazing, and 18 lowland counties (356 BBS squares), which had arable : grass ratios of 0·1–0·5 and typically consisted of more intensively managed grass leys (see the Appendix). Analyses were carried out at three complementary scales: local, regional and landscape.
To test the effect of arable habitat in grassland landscapes, the initial analyses were restricted to the 517 BBS squares in the 32 pastoral counties. The number of birds counted in each agricultural transect section in a BBS square was modelled using generalized linear models with a log–link function and a Poisson error term in SAS proc genmod (ter Braak et al. 1994; SAS Institute 1997). The number of agricultural transect sections counted (log-transformed) was treated as an offset in all models. To control for geographical trends, the northing of the Ordnance Survey grid was included in the models if it accounted for a significant change in the deviance of the model. Easting was also considered initially, but never proved significant. For some species, the assumption of a mean : variance ratio of unity associated with the Poisson distribution was not well justified, i.e. the data were over-dispersed. Thus, a scale parameter (to inflate the variances) was fitted to all models using the pscale option of proc genmod. For most species the scale parameter was less than 2·5 (maximum 4·0 for house sparrow).
Terms were selected using a forward stepwise procedure, with northing entered first followed by linear and then quadratic arable terms. Each term was retained if its addition resulted in a significant decrease in model deviance. The northing term in the models is only explicitly noted in Table 3, but was selected for the same species in each of the subsequent models for local regional and landscape analyses.
Table 3. The effect of arable habitat (number of 200-m transect sections km−2) on number of birds present in a BBS square. Coefficients (1 standard error) of the arable predictor terms retained, asterisks indicate significant differences from zero. For northing, whether bird numbers are greatest in the north or south is noted (a dash indicates no significant pattern). The reduction in deviance due to northing (ΔDN) and arable (ΔDA) terms is given for the all-county analysis. The χ2 statistic tests for significant differences in the slopes between lowland and upland counties. Asterisks indicate significance: * P < 0·05, ** P < 0·01, *** P < 0·001; insufficient corn buntings occurred in upland areas for analysis. Percentage changes in populations from 1970 to 1998 (1975 for house sparrow) are taken from Gregory et al. (2000)
The extent of arable habitat regionally was assessed using the Centre for Ecology and Hydrology's land cover data set (Fuller & Parsell 1990). This classifies all 1 × 1-km squares in Great Britain into 25 land cover types, based on images from the Landsat satellite taken (mostly) during the summer and winter of 1990. This includes an estimate of the amount of tilled land in each square. The percentage cover of tilled land in the pastoral counties was closely correlated with that recorded by BBS surveyors in the agricultural squares, despite an 8-year gap between the surveys (rp = 0·79, n = 1288, P < 0·0001).
Land cover data were used to estimate the percentage of arable cover in the landscape surrounding each of the 517 BBS squares in pastoral counties by defining a 121-km2 (a square 11 × 11 km) ‘region’ surrounding the central BBS square. As with the local analysis, we used a forward stepwise procedure with northing, arable in 1 km2, arable in 11 km2 and the interaction term (arable 1 km2 × arable 11 km2) entered successively into the model, each term being retained if it significantly reduced the model deviance.
At a broader scale, the widespread presence of a habitat is likely to affect the degree to which birds depend on it locally. To address this question, we analysed the differences between pastoral, mixed and arable counties in the slopes of the regressions of arable habitat on bird numbers. Differences were tested by the forced inclusion of a categorical ‘landscape type’ term (with three levels: pastoral, mixed and arable) and an interaction term (landscape type × arable habitat) in the model. For simplicity, only linear coefficients are presented; quadratic terms were rarely, and only weakly, significant.
There was much scatter in the relationships between arable habitat and bird numbers for two reasons (Fig. 2). First, relationships between bird numbers and habitat are likely to be complex: a broadly defined variable like amount of arable habitat is likely to leave considerable residual variation in bird numbers. Secondly, the BBS was designed as a mass-participation survey, with relatively simple recording methods (only two counts in a season) but a large number of samples. Consequently, within-square sampling variation is likely to be high, particularly in finer-scale analyses, such as those presented here. Despite this scatter, most of the granivorous passerines showing recent large population declines were also positively associated with the amount of arable habitat present (Table 3). Similarly, numbers of whitethroats were positively associated with the amount of arable habitat present. Numbers of the other species, the four finches, house sparrow and three insectivores, showed no increase with increasing amounts of arable habitat, whereas numbers of goldfinch and robin actually decreased significantly.
Upland and lowland pastoral landscapes differ greatly in character, both in the intensity of management and the types of habitat present. Table 3 therefore presents linear regressions for upland and lowland pastoral areas separately. The pattern of relationships was broadly similar to those for all grassland areas combined. In the majority of cases, the relationship between bird numbers and arable habitat was more positive in upland areas than in lowland areas, although in only three cases was the difference significant. That there were fewer statistically significant relationships in upland areas reflected both the smaller number of BBS squares sampled and lower bird densities (hence larger error estimates).
The response of yellowhammer, whitethroat and house sparrow numbers to the amount of arable land in a BBS square depended on the availability of arable habitat in the regional square (Fig. 3). Larger amounts of arable habitat in a BBS square were associated with an increase in bird numbers in areas where arable habitat was regionally scarce, but with a decrease in areas where arable habitat was relatively common. Similarly, the association between arable habitat and greenfinch and robin numbers was more negative in areas with lots of arable habitat. The remaining species showed no significant interaction terms. Thus, for some species, an increased amount of arable habitat in grassland landscapes was always associated with greater numbers, for others it was not, depending on the regional context.
The apparent effect of arable habitat on bird numbers differed between the county groupings (Table 4). In general, positive associations between arable habitat and bird numbers were weakest in arable counties, where arable habitat is widely available, and greatest in the grassland counties, where arable is comparatively scarce (see the Appendix). Mixed farming counties were intermediate. The significant results tended to be for those species most associated with arable farming and, as with the local analyses, numbers of the finches were less dependent on the amount of arable habitat than other seed-eaters. Amongst the non-seed eaters (except whitethroat), numbers tended to be negatively related to the amount of arable habitat present, particularly in arable areas.
Table 4. Interaction between the extent of arable land available locally and regionally. The slopes of the relationship (coefficients of a Poisson log–linear regression) between bird numbers and arable habitat (in a 1-km square) in three different county groupings (defined in Fig. 1) are given. The χ2 value tests for significant differences between the slopes (i.e. the interaction term of the regression model). Asterisks indicate significance: * P < 0·05, ** P < 0·01, *** P < 0·001
Arable habitat varies considerably in its attractiveness to birds, depending on the particular management practices employed; given the relatively broad-scale nature of this analysis it was possible to address this issue only rather crudely, however, it was useful in illustrating the potential effects. Thus, for all seed-eating species that were significantly related to the amount of arable habitat, the association was greater for spring cereal than winter cereal, although the differences were generally small (Table 5). Similarly, a number of species occurred in greater numbers in BBS squares that had some stubble fields present.
Table 5. The effect of arable habitat management on numbers of birds in agricultural habitats in BBS squares in grassland regions. For time of sowing, the slope of the regression (on a log scale) between bird numbers and amount of cereal habitat (± 1 std error) is given; asterisks indicate slopes significantly different from zero. For stubbles, the mean number of birds per transect section is given, with 95% confidence limits in parentheses; asterisks indicate significantly different numbers when stubble is present. Significance: * P < 0·05, ** P < 0·01; *** P < 0·001
In recent years, much has been made of the increasing intensity of management of arable systems and its impact on birds throughout Europe (Tucker 1997; Krebs et al. 1999; Aebischer et al. 2000). However, the analyses presented here show that, in grassland landscapes, increasing the amount of arable cultivation is likely to be beneficial for granivorous bird species, particularly those that have experienced severe population declines. As the availability of arable habitats in the surrounding region and in the wider landscape increases, the positive association between bird density and the local availability of arable habitat tends to weaken and in some cases is reversed. Where the latter occurs, this indicates that the presence of some grassland may be important for the persistence of some bird species in landscapes dominated by arable cultivation.
These results demonstrate the scale-dependence of bird–habitat associations, which are best shown in data sets of large geographical extent but small ‘grain’, such as the BBS. Birds are likely to select habitats hierarchically, so habitat associations will vary with scale (Kotliar & Wiens 1990). Small-scale analyses have shown that scarce features are more important in determining territory location. Thus, for example, nuthatch Sitta europaea territory occupation is correlated with the number of oak trees in coniferous woods, where they are scarce, but not in mixed deciduous woods, where they are common (Burkhardt, Schlund & Stauss 1998). This study reveals further the importance of such scale-dependence with respect to regional populations. Thus, studies of habitat associations need to consider regional context, as their strength, or even direction, may vary (Kotliar & Wiens 1990; Steele 1992).
These results support the hypothesis that range contractions (i.e. local extinctions) of some granivorous species (e.g. grey partridge, tree sparrow and corn bunting) in grass-dominated landscapes are a consequence of reduced arable cultivation. Species that have not suffered range contractions (e.g. chaffinch, greenfinch and linnet) do not seem to be strongly associated with the presence of arable habitat. This has important implications for conservation planning because it suggests that the loss of arable habitat where it is scarce may be causing declines in some areas, even though intensification of arable management is the main cause of decline elsewhere. Such distinctions present challenges to those developing agri-environmental policies adopted at the national level; very different strategies may be required in different areas.
Inspection of Fig. 2 shows that, although some species in grass-dominated landscapes tend to be found only where arable is locally common (e.g. corn bunting), others (e.g. yellowhammer, skylark and whitethroat) increase even if only small amounts of arable habitat occur within a BBS square. In western Britain, average farm size, excluding non-cropped areas such as wood and farm buildings, is around 40–50 ha (MAFF 1998). On the basis of the results presented here, if even a single field (typically 5–10 ha or c. 15% of the cropped area) on each farm was used for arable crops, farmland bird numbers could be increased.
For the granivorous species considered here, arable cultivation in grassland landscapes is likely to be beneficial through the provision of additional food sources. The major part of the winter diet of these species is the seed of arable weeds and spilt cereal grain that are available on post-harvest stubbles (Wilson et al. 1999). The annual weeds typical of arable systems are scarce in pastoral areas, particularly the highly fertilized, heavily grazed pasture common over much of lowland Britain (Wakeham-Dawson et al. 1998). A number of studies have shown that these birds prefer to forage in areas where seed density is high (Wilson, Taylor & Muirhead 1996; Robinson & Sutherland 1999). Thus, winter habitat management is likely to increase summer breeding numbers, especially for relatively sedentary species such as the yellowhammer and skylark.
None of the finches occurred in greater numbers in areas with more arable habitat. In general, finches prefer to feed on the seed of perennial weeds more associated with grassland (Wilson et al. 1999). Numbers of only one of the reference insectivores, the whitethroat, increased with the availability of arable habitat in grassland landscapes. Whitethroats use rough grassy margins and ditch banks as nesting habitat (Stoate 1999), whereas the other three insectivorous species nest in woody vegetation. Arable fields are more likely to provide this habitat because of the absence of stock grazing to the hedge base, and the difficulty of operating farm machinery in the field margins. Some of the granivorous species, notably the buntings and grey partridge, also use rough herbaceous field margins for nesting (Potts 1986; Bradbury et al. 2000; Brickle et al. 2000) and are likely to benefit in the same way.
The analyses presented here treat both arable land and agricultural grassland as though they were homogeneous habitat types. In reality, both arable and grassland vary greatly in structure, species composition and management, both within and between fields. Much recent research has shown the complexity of the mechanisms linking this variation to the species, distribution, abundance and demography of birds present (Aebischer, Green & Evans 2000). Managing arable habitat to provide over-winter stubbles, spring-sown cereals or grass margins would increase numbers of birds. Siriwardena et al. (2000) analysed the frequency of occurrence of birds breeding on British farmland and found that arable–pastoral heterogeneity per se generally did not increase bird abundance, but that particular agriculture practices were important for particular species. Further work is required to determine whether the greater importance of arable habitat in grassland habitats is because of differences in habitat management, or whether there is an interaction with the surrounding habitat. For example, pastoral farmers are unlikely to manage arable habitat as intensively as arable farmers and territories may be smaller where both grass and arable habitats are available contiguously.
The importance of over-winter stubbles has been clearly demonstrated for the cirl bunting Emberiza cirlus, whose UK population had fallen to 118 pairs, largely concentrated in south-west England. Provision of weedy stubble habitat through agri-environment schemes increased cirl bunting numbers to around 500 pairs in less than 10 years (Aebischer, Green & Evans 2000). The results presented here suggest that similar action in other pastoral landscapes could be beneficial to avian biodiversity more generally, by providing increased nesting habitat and winter food resources.
In Europe several schemes are designed to mitigate the environmental impacts of agriculture, including the Environmentally Sensitive Area (ESA) and the Less Favoured Area (LFA) schemes, which are currently being re-implemented through the Agenda 2000 reforms (Baldock et al. 2000). In addition, there are schemes specific to individual European Union (EU) member states, such as Countryside Stewardship in England. Their aim is to maintain and increase the wildlife and landscape amenity of areas under the agreements and they are largely focused on pastoral areas, where farming tends to be less profitable (Ovenden, Swash & Smallshire 1999). We suggest these schemes could be used as a basis for encouraging arable cultivation in pastoral areas; similar proposals are included in the new Welsh LFA regulations to encourage a more mixed farming landscape (Baldock et al. 2000). However, the scale of the benefits in terms of bird numbers will depend critically upon the specific management practices employed; monitoring the effect of their implementation should be a priority.
More generally, this study indicates bird species diversity and abundance are likely to fall wherever agricultural landscapes become homogeneous, whether they are dominated by intensive grass or arable management. ‘Pockets’ of arable management in grassland landscapes, and of grassland within intensively arable land, may be of great importance in allowing populations to persist at the landscape scale. The encouragement of regionally rare broad categories of farming may be of value as a general principle in agri-environmental planning.
We thank Stephen Baillie, Dan Chamberlain, Steve Freeman, David Noble, Will Peach and Gavin Siriwardena for much useful advice and for commenting on the manuscript. The three referees and Steve Ormerod also made a number of useful comments. The Breeding Bird Survey is organized by the British Trust for Ornithology and funded jointly by the British Trust for Ornithology, the Royal Society for the Protection of Birds and the Joint Nature Conservation Committee (on behalf of English Nature, Scottish Natural Heritage, Countryside Council for Wales and Environment and Heritage Service in Northern Ireland). We also thank the many volunteer fieldworkers and regional organizers for their hard work in collecting the data.
Received 7 July 2000; revision received 21 June 2001
Table 6. Table showing different types of cultivation as a percentage of the total agricultural area for the county groupings in Fig. 1.