Testing agri-environment delivery for farmland birds at the farm scale: the Hillesden experiment


Corresponding author.
Email: sahi@ceh.ac.uk


The Hillesden experiment, established in 2005/2006 to test the delivery of biodiversity benefits under Environmental Stewardship, covers c. 1000 ha of arable farmland in central lowland England. It is a randomized block experiment with five replicates of three treatments: (1) CC: cross compliance, the control; (2) ELS: 1% of land removed from production for wildlife habitat provision; and (3) ELS-X: 5% of land used for wildlife habitat, each treatment being applied to contiguous areas of 70–80 ha. Bird usage of winter food patches, comprising three different seed mixes, was monitored through the winter and was also related to seed yield. Winter and breeding season bird/territory abundance was recorded before and after the provision of the winter food patches. Bird use of the patches differed between seed mixes. There was large variation between individual patches in both seed yield and bird numbers and between individual bird species in their use of different seed mixes, suggesting that the availability of a range of patch types would be beneficial. Use of all patch types declined sharply in late January to February, indicating depletion and/or inability of birds to access shed seed. Winter bird abundance at a farm scale for all species combined, granivorous species and nine individual species increased for all monitored species when seed patches were available. At a treatment level, the increases tended to be greater in ELS-X, where most of the patches were located. In the breeding season at a farm scale, the numbers of territories for all species combined and granivorous species increased significantly when seed patches had been available in the previous winter. There was little evidence of a treatment-scale response. The provision of winter food appeared to increase winter bird abundance and to follow on into an overall increase in the breeding population, but if the latter effect is to be reflected elsewhere, this requires that sufficient breeding habitat is available to accommodate an increase.

Agriculture is the dominant land use in Europe and has a concomitant major impact on biodiversity (Pain & Pienkowski 1997, Schifferli 2000, Robinson & Sutherland 2002, Donald & Evans 2006, Voříšek et al. 2009). Low-intensity agriculture can create fine-grained landscapes with high habitat and structural diversity, but in the last 40–50 years agricultural intensification has been responsible for severe declines in many bird species, especially those most closely associated with cropped land (Fuller et al. 1995, Chamberlain et al. 2000, Fuller 2000, Hole et al. 2002, Newton 2004). Agri-environment schemes (AES) are a key component of UK Government policy aimed at mitigating habitat loss and degradation in farmland, specifically to reverse the decline in farmland birds (Swash et al. 2000, Bradbury et al. 2004, Vickery et al. 2004). The Farmland Bird Indicator (FBI; Defra 2010a), one of 18 UK Biodiversity Indicators, averages the population trends of 19 species of farmland birds (both generalists and specialists) as a means of monitoring the effectiveness of measures to improve UK agriculture for biodiversity (Gregory et al. 2004). The FBI was adopted in the late 1990s with a Public Service Agreement (PSA) to reverse the declining trend by 2020 (Defra 2008).

Currently, around 63% of agricultural land in England is under some form of agreement, at a cost of about £341 m per annum (Defra 2010b). The most recent scheme, Environmental Stewardship (ES), was made available to all farmers in England in 2005, after pilot studies in 2003/2004. By the end of 2008, some 5 million hectares were being managed within the Entry Level Scheme and 291 000 ha under the Higher Level Scheme (Natural England 2009). Despite these efforts, the overall trend in the FBI is currently downwards. Various reasons related to key ES options (and others, see Discussion) have been suggested for this continued decline, but given the level of uncertainty, there is a clear need to investigate the effectiveness of ES options as they are applied on farmland (Davey et al. 2010).

The Hillesden experiment was designed to evaluate and demonstrate the cost-effectiveness of ES options in conserving and enhancing farmland biodiversity at a farm scale and to inform future development of ES prescriptions. The work focuses on impacts across a number of taxa including arable weeds, seed/berry production, pollinators (bumble bees, solitary bees and their parasites, Carvell et al. 2008), soil micro and macro fauna, butterflies and moths, other invertebrates, small mammals and birds. In this paper, we describe results from the first 2 years of the 5-year project detailing bird responses to the provision of winter food patches of several different seed mixes and the apparent consequences for subsequent numbers of breeding territories. From the results of previous studies (e.g. Boatman et al. 2000, Henderson et al. 2004, Stoate et al. 2004), we hypothesized that food patches would attract greater numbers of birds than cropland, thus increasing bird abundance in winter and providing the potential to enhance the size of the breeding population. We examined responses of both individual species and groups of species and assessed how spatial scale and variation in the data affected our ability to detect differences between treatments compared with changes at a whole-farm scale. We also investigated the effects of patch seed yield on bird usage, with an expectation of a positive relationship, and evaluated the use made by birds of other types of prescribed margin and patch habitat types in late winter.


Study site

The experiment is located on about 1000 ha of lowland arable farmland in central England (51°57′N, 1°00′W) near Buckingham (Fig. 1). The site is a typical heavy land farm with a simple rotation of autumn-sown winter wheat, oil-seed rape and field beans. The study was established in 2005/2006 and is a randomized block experiment with five replicates of three treatments, each treatment being applied to contiguous areas of 70–80 ha. The three treatments are defined as follows:

Figure 1.

 Map of the study area showing arrangement of the replicate blocks (B1–B5) and the three treatments (CC, ELS and ELS-X) within each block. Blue lines show the locations of the 15 hedgerow transects and the black dots the 20 bird food patches. The insert shows the location of the study area in central lowland England.

  • 1 Cross compliance (CC): annual post-harvest hedge cutting and 6-m-wide buffer zones to protect hedges and water courses; this is the control treatment reflecting minimum farming environmental requirements.
  • 2 Entry-Level Scheme (ELS): 1% of land removed from production to create a small number of simple options, namely one winter bird food patch and some grass margins (6–8 m), with hedges cut every 2 years.
  • 3 Entry-Level Scheme Extra (ELS-X): 5% of land removed from production to create a more diverse range of options, including three sorts of bird food patch and a range of grass and flower margins (6–8 m) and patches, plus biennial hedge cutting.

In August 2007, airborne remote sensed data (Light Detection and Ranging, LiDAR, and hyperspectral data) were acquired for an area of c. 21 km2 centred on the study site. These data have been used to construct a high-resolution digital canopy height model (LiDAR) and a land-use map (hyperspectral), both with a pixel size of 0.5 × 0.5 m. Further details of the approach and methodology can be found in Hinsley et al. (2002, 2008), Hill and Thomson (2005) and Hill and Broughton (2009). The canopy height model was used to extract structural data describing hedgerow and tree characteristics (see below) and the land-use map is referred to later in the Discussion in relation to future work.

Winter bird food patches

The winter bird food patches were first sown in the spring of 2006, with one c. 0.25 ha patch in each ELS Treatment and three c. 0.5 ha patches in each ELS-X Treatment, giving a total area of 8.33 ha in 20 patches. To mimic realistic farm management practice, the locations of the patches within each treatment (Fig. 1) were selected by the farm manager. Each of the three ELS-X Treatment patches comprised a different seed mixture, referred to hereafter as Biennial, Bumblebird and Deluxe; the details of plant species and sowing rates are given in Table 1. Bumblebird and Deluxe were sown annually, but the Kale Brassica oleracea and Fodder Beet Beta vulgaris in the Biennial mix produced seed in their second year, following seeding by triticale (wheat/rye hybrid) and Quinoa Chenopodium quinoa in the first year. The ELS Treatment patches were all sown with the Biennial mixture. One of the ELS Treatment patches failed to establish and hence the sample sizes of the different mixtures were nine, five and five for Biennial, Bumblebird and Deluxe, respectively. In 2007/2008, the Biennial patches were in their second year, i.e. the Kale and Fodder Beet were in seed.

Table 1.   Composition at sowing of winter bird food patches. The Biennial mix (ELS and ELS-X) was sown at 40 kg/ha, the Deluxe (ELS-X) and Bumblebird (ELS-X) at 20 kg/ha.
40Triticale (wheat/rye hybrid)40Triticale70Triticale
15Millet (Echinochloa esculenta)20Millet14Kale (Brassica oleracea)
15Dwarf Sunflower (Helianthus annuus)20Buckwheat (Fagopyrum esculentum)14Quinoa
10Quinoa (Chenopodium quinoa)10Quinoa2Fodder Beet (Beta vulgaris)
10Fodder Radish (Raphanus sativus)10Fodder Radish  
7Sweet Clover (Melliotus officinalis)    
3Borage (Borago officinalis)    

Maximum seed production was estimated from samples taken from the patches during September each year. In each of the patches, five quadrats (1 m × 0.5 m) were sampled arbitrarily, avoiding the first 3 m from the edge, but utilizing the whole patch. The number of individuals of all plant species (i.e. sown species and weeds rooted in a quadrat) were counted and recorded, along with their reproductive status (seeding/non-seeding). All seeds and seed heads from each of the seeding species were collected and stored separately in labelled plastic bags. If samples were not processed immediately, they were frozen. When processed, the seed from each species was dried at 80 °C for 24 h and then weighed.

Bird use of the bird food patches, and relative use of each patch type, was monitored throughout the winter in 2007/2008. Counts of birds, identified to species, present in all 20 sown patches were made on each of six visits (October, November, December, early January, late January and February). To minimize the risk of multiple counts of birds moving between patches, all patches were counted on the same day at approximately the same time by five observers each counting four patches. Patches were observed at a distance and eventually flushed by walking around the patch perimeter and then through the patch. The aim of the counts was to record the total number of birds in each patch. Counts took c. 20 min to 1 h to complete, depending on bird numbers and species composition and the vegetation density of the patch. To obtain a comparison with bird use of cropland, each patch was paired with an equivalent area of crop, with a similar shape and location, in an adjacent field, which was counted immediately after the patch count.

Late winter use of other patch and margin habitats

Three counts (not reported here) of the bird food patches made in the previous winter (i.e. 2006/2007), and results from other studies (e.g. Siriwardena et al. 2008), indicated that bird usage declined in late winter (late January/February, see below). Thus potential bird use of alternative prescribed habitats, i.e. margins and other patch types, was also monitored in January to the beginning of April in 2008. Samples of different margin and patch types, including bird food patches, located chiefly in the ELS-X Treatments, were counted 12 times, the observer walking at a steady pace and recording all birds seen or flushed from each habitat type. The habitat types and the total area of each sample are given in Table 2. The numbers of birds encountered in all habitats except the bird food patches were small and thus results were expressed as the numbers of birds per habitat type, and numbers per unit area, totalled across all 12 visits.

Table 2.   Extent and use of different patch and margin habitat types in late winter (January to first week of April, 2008).
Habitat typeTotal area recorded (ha)Total no. of birds countedBirds per 100 m2
All speciesGranivorous species
  1. EF1 (management of field corners: creation of grass and wildflower patches, 0.5 ha, in field corners that are awkward to reach with machinery) and EF2 (wild bird seed mixture patches: mixtures of annual and biennial small seed-bearing crop species sown in low yielding or awkward patches, 0.25–0.5 ha) are ES habitat designations (Anon 2009).

Bird food EF2 patch4.8411442.361.81
Tussocky margin1.341210.900.72
Pollen and nectar margin2.29610.270.11
Flower EF1 patch3.64450.120.04
Annual cultivation margin1.28100.080.02
Natural regeneration margin0.7090.130.00

Winter bird population census

The birds present in winter were also recorded at the scale of each treatment (i.e. CC, ELS and ELS-X) in each of the five replicate blocks. Birds were monitored using hedgerow transects located in the interior of each treatment (with one exception where c. one-third of the transect in one CC Treatment comprised a treatment boundary; Fig. 1). Within each treatment, the hedges were usually contiguous, with a few instances of interspersed small copses, clumps of bushes/trees and gaps. Selected structural details of the 15 transects are given in the supplementary material (Table S1). All birds in the hedges, hedgerow trees and within 10 m of either side of the area were recorded by spot mapping. Bird locations and activities were recorded on large-scale maps which included the locations of all hedgerow trees and other landmarks. Birds in the transects were recorded on three visits (November, December and January) in both of the winters of 2005/2006 and 2006/2007, i.e. before and after bird food patch establishment. Counts began a little after dawn and finished at least 2 h before dusk; counts were not undertaken in weather likely to depress bird activity, such as rain or high winds. Each count was usually completed within 2 days, or occasionally 3 days in the event of poor weather. The bird records were later digitized using the LiDAR canopy height model within a geographical information system (ArcMAP v9.3 © 2008 ESRI Inc.). The birds recorded in the winter comprised a mixture of residents, small- and regional-scale migrants and long-distance migrants in unknown proportions. Therefore, the overall winter abundance of each species was expressed as the number of individuals per kilometre per transect totalled across all three counts. Totals (per km) were also calculated for each transect for ‘all species’ and for ‘granivorous species’ (Chaffinch Fringilla coelebs, Linnet Carduelis cannabina, Yellowhammer Emberiza citrinella, Reed Bunting Emberiza schoeniclus, Tree Sparrow Passer montanus).

Breeding bird territory census

The breeding birds within each treatment were recorded by territory mapping using the same 15 hedgerow transects (Table S1) used for the winter bird counts. Birds, and their activities, in the hedges and hedgerow trees and within 10 m either side were mapped as in the winter. Particular attention was paid to behaviour indicative of breeding. Four visits (April, May, June and early July) were made in the breeding seasons of 2006 and 2007, i.e. before and after the availability of the winter bird food patches. The bird records were again digitized and territory locations were estimated using observer judgement based on recorded bird behaviour to assign each record to a territory (Bibby et al. 1992), and then generating minimum convex polygons to represent approximate territory boundaries. A 10 m buffer was placed around the resultant polygons to account for possible error when census mapping and digitizing locations. Territories estimated to occur wholly or substantially (more than two-thirds minimum convex polygon area) within each treatment were counted for each species. The total numbers of territories per kilometre per transect were then calculated for each individual species and also for the two groups of ‘all species’ and ‘granivorous species’.

Data analysis

To account for non-normality and the non-negative integer property of the data, a Poisson modelling approach was used. Where the fit of Poisson models was poor, P values were corrected using the Quasipoisson adjustment, all models being run in R 2.10.0 (R Development Core Team 2009).

To examine differences in bird numbers (for all species combined and for individual species) between patch types for each visit, generalized linear models with a Poisson distribution for the bird counts and log-link function were used, with log (patch area) as an offset to allow for differences in patch area. Patch type (Biennial, Bumblebird, Deluxe) was used as a predictive factor plus length of woody boundary to accommodate any effects of patch boundary structure. Models fitting individual slopes for each patch type were compared with those fitting a common slope.

The relationships between individual patch seed production, measured in September (kg of seed per patch, all plant species) and the bird counts (all species combined) from October to February were examined using generalized linear models with a Poisson distribution for the bird counts, patch type as a factor plus seed yield and an interaction between type and seed yield. When the interaction was not significant, models using a single slope and different intercepts for patch type were compared with those fitting a single slope and intercept. Area was not included as an offset because the response being modelled was the observed response of the birds to the actual seed yield of each patch.

Differences in bird numbers both in the winter and during the breeding season in the years before and after bird food patches were available were compared at the whole-farm scale using generalized linear models with a Poisson distribution for the bird counts (all species combined, granivorous species and individual species) for each transect. The influence of Chaffinch, the most numerous species in both seasons, was also examined by removing numbers of Chaffinches from the totals for ‘all species’ and ‘granivorous species’. Differences between years were assessed by comparing models including transect and year as factors with models omitting year effects. To examine treatment-level (CC, ELS, ELS-X) effects on differences between years for ‘all species’ and ‘granivorous species’, the models were re-run including treatment as a factor and an interaction between treatment and year. Models with and without the interaction were then compared. Where treatment had a significant effect, differences between years were examined separately for each treatment using transect and year as factors.


Winter bird food patches

As expected, the abundance and species richness of birds counted on the bird food patches were substantially higher than those recorded in equivalent areas of crop. In 2007/2008 (Biennial mixture in its second year), the mean number of birds and the mean species richness per patch per visit (all patches combined) standardized to an area of 1 ha were 79 and 8.7, respectively, compared with 0.3 birds and 0.2 species in 1 ha of crop. Considering PSA species only, these figures were 27 birds and two species in patches compared with 0.06 birds and 0.02 species in crops. The PSA species recorded in patches were Linnet, Goldfinch Carduelis carduelis, Greenfinch Carduelis chloris, Yellowhammer, Reed Bunting, Skylark Alauda arvensis, Grey Partridge Perdix perdix and Kestrel Falco tinnunculus, and those recorded in the crop were Skylark and Grey Partridge.

The numbers of birds counted varied between patch types (Fig. 2), but due to large variation in the numbers present on individual patches (see below), the only significant difference occurred in November (F2,16 = 8.587, = 0.003, Fig. 2). There was no effect of length of woody boundary in any month. Overall, the Bumblebird mixture attracted fewer birds and, in general, bird numbers on all patch types declined in late winter (late January, February). There was large variation in the responses of individual species to the different patch types; examples are shown in Figure 3. Apart from an initial presence in Deluxe patches in early winter, Yellowhammers were rare, whereas Linnet numbers were maintained in Deluxe until late winter (Fig. 3a,b). Song Thrushes Turdus philomelos showed an increasing trend in the Biennial mix until late winter (Fig. 3c) and Dunnocks Prunella modularis occurred in all three patch types (Fig. 3d). Chaffinches (Fig. 3e) showed a similar pattern to that for all species combined (Fig. 2), whereas Greenfinches (Fig. 3f) showed a similar response to Yellowhammers to Deluxe and a peak in numbers in the other two types in December. Goldfinches (Fig. 3g) were commonest in the Biennial mix, declining in numbers here throughout the winter, but were, at least in part, attracted by Chicory Cichorium intybus, which had occurred as an accidental contaminant. Reed Buntings (Fig. 3h) occurred sporadically in low numbers, but were absent from the Biennial mix apart from a single bird in one patch in early January. As with the results for all species combined, the large variation in counts of individual species between individual patches caused most differences between patch types to be either non-significant or to occur sporadically in different months for different species (Table 3). The exceptions were the increasing and/or decreasing trends for Song Thrush, Chaffinch and Goldfinch (Fig. 3c,e,g). A significant boundary effect was found on three occasions (Yellowhammer, October, F1,16 = 11.249, = 0.004; Greenfinch, October, F1,16 = 6.295, = 0.024; Reed Bunting, October, F1,16 = 10.588, = 0.005) and implied a negative relationship between bird numbers and the length of woody boundary. However, for Yellowhammer and Reed Bunting, birds were only present in two and three patches, respectively, and the numbers of the latter species were low (one, two and five birds per patch). Therefore, these results should be treated with caution, and additional data to elucidate possible boundary effects are currently being collected.

Figure 2.

 Bird usage of the three different types of bird food patch during the winter of 2007/2008. Bird numbers (for all species combined) are mean counts per visit date per patch type, corrected to a standard patch area of 0.5 ha. Date 0 = 1 October; error bars show + 1 se only for clarity.

Figure 3.

 Examples of individual bird species usage of the three different types of bird food patch during the winter of 2007/2008. Bird numbers are mean counts per visit date per patch type, corrected to a standard patch area of 0.5 ha. Date 0 = 1 October; error bars show + 1 se only for clarity.

Table 3.   Differences (illustrated in Fig. 3) in the use of the three types of bird food patch by individual bird species during the winter of 2007/2008.
SpeciesOctoberNovemberDecemberEarly JanuaryLate JanuaryFebruary
  1. Results are shown for significance levels up to = 0.10 (plus one additional result for Chaffinch); for all other results, P exceeded 0.10. Note that results for Reed Bunting should be treated with caution due to small numbers of birds (see Fig. 3h).

Song Thrush
 P0.0310.0050.0820.066< 0.0010.083
 P0.010< 0.0010.0230.0890.1290.074
 P0.010< 0.0010.108
Reed Bunting

A positive relationship between the amount of seed produced by a patch and the total numbers of birds present at the beginning of the winter was expected. However, the relationship was significant only in November (slope = 0.00152, F1,18 = 8.297, = 0.010, Fig. 4), the best model using a single slope and intercept. There was no evidence of a relationship in either October (slope = −0.00030, F1,16 = 0.095, = 0.762) or December (slope = 0.00127, F1,16 = 1.212, = 0.288). This was probably due to the large variation in seed yield between patches, including within the same type, and also between bird counts on individual patches. Relationships between seed yields and bird counts later in the winter were not expected (due to the length of time from measurement of the seed crop in September) and none was found (Early January: slope = −0.00044, F1,16 = 0.090, = 0.768; Late January: slope = −0.00013, F1,16 = 0.016, = 0.901; February: slope = −0.00075, F1,16 = 0.354, = 0.561).

Figure 4.

 Effect of seed yield on bird usage of individual patches in November 2007. Fitted line calculated from Poisson model using single slope and intercept. (Note that the results include data for the failed Biennial mix patch which was resown in 2007 with the Deluxe annual mix; no seed crop data were available for one of the Bumblebird patches.)

Late winter use of other patch and margin habitats

The largest numbers of birds were recorded on the bird food patches (Table 2) with only small numbers recorded on any of the other margin or patch habitat types. The most frequent species on tussocky margins were Goldfinch and Reed Bunting, the former being attracted to Teasels Dipsacus fullonum. In pollen and nectar margins, partridges (mostly Red-legged Partridge Alectoris rufa) and Chaffinch were the most frequent species. No particular species were associated with any of the other habitat types, and in all cases, the numbers of birds present was small (Table 2).

Winter bird population census

At a whole-farm scale, using the 15 transects as replicates, more birds were recorded in the winter of 2006/2007, following bird food patch establishment, than in the winter before (2005/2006) (Table 4). Numbers were greater after establishment for all the species recorded and the differences were significant for the two groups of ‘all species’ and ‘granivorous species’ and for nine of the individual species (Table 4). The increase between years for ‘all species’ and ‘granivorous species’ remained significant when Chaffinch was excluded from the groups (all species: F1,14 = 6.692, < 0.001; granivorous species: F1,14 = 6.810, < 0.001). At the treatment-level, for ‘all species’ and ‘granivorous species’ numbers increased after establishment in the form of ELS-X > ELS > CC (Fig. 5a), but treatment was significant only for ‘granivorous species’ (F2,12 = 6.483, = 0.012). For ‘granivorous species’, the difference between years was significant for all three treatments (CC: F1,5 = 17.009, = 0.015; ELS: F1,5 = 36.161, = 0.004; ELS-X: F1,5 = 92.740, < 0.001; Fig. 5a).

Table 4.   Winter bird abundance (mean numbers of birds per kilometre of transect, = 15) at a farm scale before (2005/2006) and after (2006/2007) the establishment of winter bird food patches and other habitats.
SpeciesWinterF1,14P value
All species52.4118.55.701< 0.001
Granivorous species9.951.03.370< 0.001
Blackbird Turdus merula10.010.80.0680.798
Chaffinch Fringilla coelebs8.427.74.336< 0.001
Robin Erithacus rubecula7.78.20.1050.751
Blue Tit Cyanistes caeruleus7.313.114.0220.002
Song Thrush Turdus philomelos4.68.05.8570.030
Dunnock Prunella modularis4.413.245.022< 0.001
Wren Troglodytes troglodytes4.45.61.5160.238
Great Tit Parus major2.75.39.0020.010
Yellowhammer Emberiza citrinella1.311.866.799< 0.001
Linnet Carduelis cannabina0.27.414.8440.002
Tree Sparrow Passer montanus0.10.728.186< 0.001
Reed Bunting Emberiza schoeniclus0.03.44.518< 0.001
Figure 5.

 Differences in bird abundance (for all species combined and granivorous species) between treatments in the winter: (a) mean number of birds per km of transect, and breeding season: (b) mean number of territories per km of transect – before (2005/2006 and 2006) and after (2006/2007 and 2007) bird food patch establishment. CC = Cross Compliance, ELS = Entry Level Scheme, ELS-X = Entry Level Scheme Extra; error bars show ± 1 se.

Breeding bird territory census

More breeding territories were recorded in the breeding season (2007) following bird food patch establishment than in the previous breeding season (2006) (Table 5). Numbers of territories increased after establishment (albeit marginally for some species) for all the species recorded except Reed Bunting (resident species) and Whitethroat Sylvia communis (summer migrant) and the differences were significant for ‘all species’, ‘granivorous species’, Chaffinch, Dunnock and Robin, and nearly so (= 0.053) for Linnet (Table 5). The significance for ‘granivorous species’ was due to Chaffinch (without Chaffinch: F1,14 = 2.354, = 0.125), but the significance for ‘all species’ was independent of Chaffinch numbers (without Chaffinch: F1,14 = 5.390, = 0.020). There were no significant differences at the treatment level; unlike the winter data, the pattern of numbers across the three treatments was similar in both years for both groups of species (Fig. 5b).

Table 5.   Breeding season bird abundance (mean numbers of territories per kilometre of transect, = 15) at a farm scale before (2006) and after (2007) the establishment of winter bird food patches and other habitats.
SpeciesBreeding seasonF1,14P valueNational trends (%)
  1. Results are compared with the percentage change from 2006 to 2007 in national populations (Risely et al. 2008).

  2. *Significant change.

All species33.939.98.1760.004
Granivorous spp.
Blue Tit2.83.62.9450.108−6−7
Great Tit2.32.60.6530.4330−5
Reed Bunting1.21.10.0280.869−2+1
Song Thrush0.60.70.5810.459+1+1


It was no surprise that many more birds were counted on the bird food patches than on equivalent areas of crop (e.g. Boatman et al. 2000, Henderson et al. 2004, Stoate et al. 2004, Field et al. 2009), but the winter census based on the 15 transects indicated an increase in bird numbers across the farm in general, not just on the patches. This implies that birds do not simply move between food-rich patches across the landscape, but also use intervening habitat of different types. A response to habitat other than, or in addition to, the bird food patches was also indicated by the results for Yellowhammer (Whittingham et al. 2005) and Blue Tit. Both these species showed a significant increase in numbers in the winter of 2006/2007 (Table 4), but neither was present in large numbers on the bird food patches (Fig. 3). Overall, the species showing the greatest response to the patches in winter were, as expected, the seed-eaters, plus Song Thrush and Dunnock (see below). With the exception of Blue Tit (which appears to be less sensitive to habitat quality than Great Tit; Hinsley et al. 1999), woodland or woodland edge species such as Robin and Wren tended to show little response.

The increase in the numbers of breeding territories in 2007, following patch establishment, also suggested that the birds were responding at a whole-farmscale and that, given the availability of suitable habitat, had been encouraged to remain in the area to breed (Gillings et al. 2005). The fact that most of the trends in both winter and breeding season bird numbers were positive (Tables 4 & 5) suggested a positive effect of the overall habitat provision under ES. The general increase in the numbers of breeding territories at Hillesden from 2006 to 2007 contrasted with the changes in the national English population monitored by the BTO/JNCC/RSPB Breeding Bird Survey (Risely et al. 2008, Table 5). At the national scale, six species showed little change or an increasing trend from 2006 to 2007, whereas seven, including Linnet and Yellowhammer, showed a declining trend. This, together with the fact that there was no other major habitat alteration on the farm between the two winters and that the bulk of the census work was carried by the same two observers in all years, suggested that there was a genuine positive response to the ES habitat management.

Bird exploitation of dedicated food patches will depend on many factors including seed type and yield, patch location, and the distribution and abundance of alternative food supplies in the wider landscape (Whittingham & Evans 2004, Siriwardena & Stevens 2004, Stoate et al. 2004, Siriwardena et al. 2006, Siriwardena 2010). Some species, such as Dunnock and Song Thrush, may also respond to attributes other than those directly related to the sown seed crop such as shelter, access to damp ground, and associated weed seed and invertebrate food resources (Peach et al. 2004, Gilroy et al. 2008). This was thought to be especially pertinent for Song Thrush (and other thrushes) in the tall dense cover offered by the Biennial patches (Fig. 3, Table 3). Similarly, the presence of Linnets in the Deluxe patches (Fig. 3b) was thought to be due in part to growth of weeds (e.g. Chickweed Stellaria media and Groundsel Senecio vulgaris) and, as mentioned above, Goldfinches responded to the accidental presence of Chicory in the Biennial mix (Fig. 3g, Table 3). The range of responses of individual species to the different seed mixtures used at Hillesden indicated that selecting a single ‘best’ mix was unlikely to be feasible; different bird species are well known to prefer different seed types and sizes (Wilson et al. 1999, Boatman & Stoate 2002, Stoate et al. 2004, Holland et al. 2006). There is also the choice of annual vs. biennial mixes; the results for the Biennial mix reported here were dominated by the characteristics of the second year seeding species, Kale and Fodder Beet. Therefore, provision of a range of patch types, within a single farm or co-ordinated across larger areas, could be most cost-effective.

There is also the possibility that particular bird species could be targeted in particular locations. Trials of such an approach are being carried out at a regional scale (Phillips et al. 2009), but there is no reason why targeting could not take place at a farm or even field scale to benefit certain local populations, e.g. a colony of Tree Sparrows. In addition, the physical attributes of the patches differed substantially, e.g. in terms of sown species heights and densities, and this variation was increased enormously by additional variation in growth performance and seed production of individual patches (Fig. 4). Bird food patches, and other non-crop habitat types, tend to be located in the least favourable/least productive areas of fields, and this, coupled with the repeat sowing of the same areas across several years, can result in poor establishment and performance. Thus the quality of ES options is likely to be as important as their identity; the production and maintenance of good quality bird food, and other semi-natural habitat will probably require a similar management effort to that devoted to crops (Stoate et al. 2004, Siriwardena & Anderson 2007, Douglas et al. 2009, Lobley et al. 2009).

Although a positive relationship between increasing seed yield and bird numbers was expected, this was found only in November, probably due to the large variation in both variables (Fig. 4). Many factors could have contributed to low bird counts, such as random disturbance of a patch, e.g. by a Sparrowhawk Accipiter nisus, and ‘normal’ movements of flocks between patches and other habitat. Ideally, patches could have been counted more frequently or observed for extended periods, but the result for November did indicate that more productive patches are likely to attract more birds, at least in early winter. There was an indication that the nature of the boundary of a patch might influence bird numbers, and it is well known that certain species prefer open landscapes and avoid otherwise suitable habitat in close proximity to woodland and tall hedges (Chamberlain et al. 2009). The patches at Hillesden are bordered in part by woody vegetation sometimes including lines of trees. It is possible that this degree of shelter deters some species, but equally it might attract others. The important point is that location can influence the value of ES habitat to birds (Siriwardena & Stevens 2004) and thus patches with persistently low usage are probably best relocated elsewhere. We are currently collecting more data on bird use of patches in relation to their location and boundary features to investigate species-specific responses to these factors.

Despite the large variation in the use of the three seed mix types (Figs 2–4), they all showed a sharp drop in bird numbers in late January and early February, the beginning of the so-called ‘hungry gap’ (Siriwardena et al. 2008) when both managed and natural food supplies become depleted (Hinsley et al. 2010) and/or less accessible to some species due to seed drop. Filling this gap may require novel seed crops or greater areas of dedicated bird food and there is also the possibility of ‘artificial’ feeding. However, the latter often involves the use of cereals, and such relatively large and hard seed may not be suitable for all species. Even though sown crops may be depleted in late winter, providing these additional/alternative resources earlier in the winter may prolong the availability of ‘natural’, i.e. non-cropland, food supplies. The other types of ES habitat patches and margins were little used by birds in late winter, but this was perhaps not surprising given that they are not designed to provide winter foraging habitat for birds. This does, however, strengthen the case for promoting winter bird food patches as a preferred ES option. It might also be possible to increase the value of other types of ES habitat for birds in winter by, for example, allowing grass margins to go to seed (Buckingham & Peach 2006) and including additional plant species such as Teasel. There is also the question of where the birds go in late winter. Our observations suggest that they move more widely in the landscape than the farm scale and this is consistent with various studies of winter food use (Robinson et al. 2004, Siriwardena et al. 2006) and of winter bird movements in general (e.g. Prŷs-Jones 2002). This in turn suggests that consideration of local landscape factors and co-ordination of ES management at greater than a farm scale will be necessary to maximize effectiveness for farmland birds. The problems of effective food provision in late winter in relation to bird movements in the landscape are discussed by Siriwardena (2010).

A range of possible reasons for the continuing decline in the FBI, despite the substantial investment in ES (and other AES), have been suggested. These include a lack of uptake of key options, problems with the quality, location and management of options, and the possibilities of ES benefits being offset by other (unknown) factors or being obscured by time lags in the bird population response and/or the detectability of such a response at a national scale (Chamberlain et al. 2000, Risely et al. 2008, P.V. Grice pers. comm.). Option choice, quality, location and management have been discussed briefly above. At the farm scale of the Hillesden experiment, there was no time lag in the response of bird numbers in winter to the presence of bird food patches, but the response in terms of numbers of breeding territories, although encouraging, was less clear-cut. Although the overall numbers of territories increased, the responses of individual species were more varied, and often non-significant (Table 5). Despite the contrast with the English national population trends for 2006–2007 (Table 5), a single year’s results may reflect local population redistribution rather than a general increase; detecting a long-term, sustained effect will be influenced by many events, not least the recent cold winters.

Given the caveat of the current short-term nature of the monitoring at Hillesden, and the fact that, although it is a farm-scale experiment, it is a single site, the results suggest that winter survival is a key requirement that can be addressed by providing food patches. In the second winter, there was a clear trend for bird numbers to be higher in the ELS and ELS-X Treatment areas (Fig. 5a), much as expected from the locations of the patches. With three patches, the resources in the ELS-X Treatment were greater than those of standard ELS, but the latter could be increased by simply increasing patch size. In spring, the pattern of the numbers of territories across the treatments was largely similar both before and after the patches were established (Fig. 5b), suggesting that territory distribution occurred on a larger scale than that of the individual treatment blocks. Thus winter survival appeared to be key (but see Siriwardena et al. 2000), but sufficient potential breeding habitat to support an increase in the breeding population (Siriwardena & Anderson 2007) was also implicit in the results. Smaller-scale effects, of the other ES margin and habitat types, as well as the semi-natural (hedgerow, tree, woodland, riverine, etc.) and anthropogenic (gardens, farmyards, etc.) habitats might be expected to influence the birds at the scale of individual territories. Other taxa within the experiment, e.g. solitary bees and some butterflies, could also be expected to respond to habitat on a smaller scale than that of the birds. Future analyses will use the remote sensed habitat data to investigate such smaller-scale effects, and in particular, the breeding productivity of individual bird territories.

In conclusion, the provision of winter food increased the numbers of birds present at a farm and sub-farm scale in winter and there was some evidence that this, perhaps coupled with the availability of other ES habitats (Table 2), subsequently increased the breeding population. However, settlement to breed requires that sufficient nesting/foraging habitat is available to support an increase. Thus the management of ES as a scheme must ensure the uptake of a suite of options that provide both winter food supplies (productive patches in accessible locations) as well as spring/summer nesting and foraging habitat.


We would like to thank Faccenda Farms, especially the owner Mr Robin Faccenda and the manager Mr Richard Franklin, for permission to work at Hillesden; without their help, co-operation, tolerance and interest the study would not have been possible. We also thank Defra and Natural England for planning, management and financial support, Syngenta for funding the acquisition of the remote sensed data (and additional study of the farmland tit and Tree Sparrow populations), Marek Novakowski (Wildlife Farming Company) for agronomy and management support, Ben and Jane Carpenter for winter bird census work, Peter Nuttall, Lucy and Sarah Hulmes and Jodey Peyton for vegetation sampling, and referees and editorial staff for improvements to the manuscript. Special thanks to Steve Freeman for statistical guidance and training in the use of R.