Habitat associations and breeding success of yellowhammers on lowland farmland


R.B. Bradbury (fax 01865 271221; e-mail richard.bradbury@zoology.ox.ac.uk).


1. Yellowhammers began to decline on British lowland farmland in the late 1980s and losses are presently 10% per year. This study examined variation in the habitat selection and breeding success of yellowhammers, allowing an evaluation of whether Britain's yellowhammer decline might have been caused by recent changes in agriculture.

2. Yellowhammer territories were associated with hedgerows, vegetated ditches and wide uncultivated grassy margins around fields. Pasture and silage leys were avoided. Nests were built among herbaceous vegetation in ditches or in the shrubby vegetation of hedgerows.

3. Breeding started slightly earlier on organic farms than on intensively managed farms, but no measure of breeding success differed between farm types. Predation was the cause of most (64%) nest failures. A maximum of three breeding attempts (two successful) was observed per pair, with a mean clutch size of 3·3, a Mayfield nest success rate of 0·46, and 2·6 nestlings fledged per successful brood. These data, together with published estimates of adult yellowhammer survival and of post-fledging survival among other passerines, suggest that breeding productivity is too low to maintain a stable population.

4. The removal of hedgerows or abandonment of hedge management, filling or clearing of ditches, intensification of grassland management and cropping or grazing right up to the field edge, are all likely to have adversely affected yellowhammers on lowland farmland in southern England. Policy reforms that redirect subsidy support to environmentally beneficial management of field margin habitats and retention of winter-feeding sites such as stubbles should assist in restoring populations of breeding yellowhammers on lowland farmland.

5. Our data expand further the array of farmland bird species for which interactions between agricultural change and population change are increasingly understood.


Many British farmland birds have recently undergone pronounced contractions in range (Gibbons, Reid & Chapman 1993) and declines in abundance (Fuller et al. 1995; Siriwardena et al. 1998). These declines have coincided with rapid intensification of farming practice (Pain & Pienkowski 1997) and are paralleled by declines in many of the birds' invertebrate and weed seed foods (Donald 1998; Wilson et al. 1999). Similar declines have been seen in other countries that have undergone agricultural intensification (Tucker & Heath 1994). Loss of biodiversity in agricultural environments is now considered a high conservation priority in Europe (Tucker & Evans 1997), and the UK government has committed itself to biodiversity action plans for 12 farmland bird species (United Kingdom Biodiversity Steering Group 1995).

The yellowhammer Emberiza citrinella L. is abundant over much of Europe (Cramp & Perrins 1994), preferring patchy mosaic habitats and avoiding totally tree-less agricultural landscapes or densely forested areas (Hagemeijer & Blair 1997). British farmland yellowhammers typically build their nests along field boundaries, either on the ground in ditches or grassy margins, or in hedges and isolated bushes (Parkhurst & Lack 1946; Kyrkos 1997; Stoate, Moreby & Szczur 1998). Nests on the ground are hidden in dense herbaceous vegetation. In Britain, the species is largely sedentary, wintering in the same general areas as occupied in the breeding season: 70% of ringed birds recovered in winter are within 5 km of their natal area (Lack 1986).

Although still a common breeding species in Britain, with an estimated 1200 000 breeding pairs in 1988–91 (Gibbons, Reid & Chapman 1993), yellowhammer populations began to decline in the late 1980s (Fuller et al. 1995). The rate of decline has accelerated rapidly (Siriwardena et al. 1998) with, for example, a decline of 11% between 1995 and 1996 on farmland Common Birds Census (CBC) plots, and the population is now at an all-time low since the start of the CBC scheme in 1961 (Crick et al. 1998). Similar declines have been noted in other western European countries (Tucker & Heath 1994; Hagemeijer & Blair 1997).

The aims of this study were fourfold. First, we examined the extent to which agricultural land use and non-crop habitat structure explained the distribution and breeding success of territorial yellowhammers. Secondly, we combined our data on breeding success with estimates of survival rates from national ring-recovery data sets to assess whether current levels of breeding success are sufficient to maintain populations. Thirdly, we used these results to assess whether and which changes in agricultural land use and land management practice are likely to have caused the decline in yellowhammer populations, and through which demographic mechanisms. Finally, by identifying the agricultural conditions under which yellowhammers persist at high densities and comparing them with the requirements of other well-studied species, we identified management recommendations for the conservation of yellowhammers and other declining farmland bird species.

Study sites and methods

Study sites

The study was carried out from April to July 1994–97 at nine lowland farms, in four areas within Oxfordshire, Wiltshire and Warwickshire, UK. All sites consisted of mixed farmland, with fields bounded by ditches, hedges or tree lines. Land use consisted of a combination of arable crops [autumn- and spring-sown cereals, legumes, maize Zea mays L., oilseed rape Brassica napus L. ssp. Oleifera (DC.), kale Brassica oleracea L. and linseed Linum usitatissimum L.], hay and silage for winter forage, and pasture grazed by cattle, sheep and horses. Rotational set-aside land was present either as a ‘green cover’ of weeds and crop volunteers, regenerating naturally from the stubble of the previous crop, or as a sown mixture of grass and clover (Table 1). Four farms (two in one area and one in each of two other areas) were under organic management, thus providing a range of agricultural practice across the study areas. One of the farms (site 1 in Table 1) began to convert some land to organic status during the study. The organic farms all conformed to standards of the UK Soil Association and retained many of the features characteristic of lowland agricultural systems prior to the onset of general and widespread intensification in the 1950s (Lampkin 1990). On the organic farms, there was no routine use of agrochemicals and a crop rotation including nitrogen-fixing legumes for soil-nutrient replenishment and weed control was integral to the farming regime. In the absence of pesticide use, pests were controlled mainly through this rotation system, whilst infestations of grass weeds (chiefly black grass Alopecurus myosuroides L. and wild oats Avena fatua L.) in the cereal crops were controlled by raking and cutting. The other farms (henceforth ‘intensive’) were all managed using agrochemical fertilizers and pesticides.

Table 1.  Land use composition (%), total area (ha) and farming system (or = organic, in = intensive) of the cropped area of the nine study sites. Site 1 began conversion to organic status during the study. The number in superscript indicates the ‘geographical area’ in which the site was located. Note that study areas were not whole farms. wcr, winter cereals; scr, spring cereals; leg, legumes; osr, oilseed rape; sil, silage ley; gra, grazed pasture; nrsa, natural regeneration set-aside; gsa, grass-sown set-aside; bea, beans; lin, linseed; mai, maize; kal, kale. Density1 is the number of yellowhammer territories per hectare on the farm. Density2 is the number of yellowhammer territories per 100 m of field boundary. Territory locations were identified by CBC methods. Those on the edge of a site were scored 0·5, those entirely within a site scored 1
SiteYearwcrscrlegosrsilgranrsagsabealinmaikalTotal areaDensity1Density2
11 (in)199442·32·204·019·911·910·506·203·102500·0780·075
21 (or)199428·800045·37·80010·108·00169·70·0710·132
31 (or)199442·900022·90021·412·800069·60·1650·183
42 (in)199441·90020·68·915·91·111·60000128·90·1510·174
53 (or)199625·525·90033·415·3000000197·560·0940·121
63 (in)199616·0007·950·514·4000011·2087·00·0860·049
74 (in)199667·00031·501·3000000102·390·2150·2
84 (or)199678·10007·012·802·1000056·770·2910·156
91 (in)199773·500013·510·90002·000182·460·0900·134

Habitat recording

Description of field boundaries was based on the protocol of Chamberlain, Wilson & Fuller (1999) and is detailed in Table 2. The length of each boundary section was measured from 1 : 2500 maps. Boundary units were defined as any contiguous length of field boundary between points of intersection with other field boundaries. If the nature of the boundary changed abruptly between intersections, it was further subdivided into separate recording units. Boundaries were largely unaltered during the course of the study, apart from the removal of a few dead trees and periodic winter hedge trimming, resulting in only small height variations. All boundaries at a site were surveyed in August and September of the first year in which the site was used. Thereafter, surveys were carried out at the same time of year in all subsequent years to note any changes in management.

Table 2.  List of habitat parameters recorded and used as potential predictors of yellowhammer territory density
Continuous variables
Length of boundary(m)
Number of intersections
Total woody plant species
Total number of live trees
Total number of dead trees
Maximum width of field margin(tilled fields only)
Total number of gaps > 2m wide(hedgerows only)
Categorical factorsLevels
Farming regime1 = organic, 2 = intensive
Aspect1 = north/south, 2 = east/west
Hedge, fence, tree line, ditch,0 = absent, 1 = present for each variable
wood-edge, farm-track, road
Woody species in boundaryFor every species: 1 = absent, 2 = rare, 3 = uncommon, 4 = common, 5 = abundant
Adjacent land useFor every crop/land use category: 0 = absent, 1 = present
Hedgerow boundaries only
Hedgerow height/width(m)
Hedgerow profile1 = rectangular, 2 = rounded, 3 = A-shaped, 4 = irregular
Hedgerow trimming0 = no regular trimming, 1 = regular trimming

Bird surveys

Yellowhammers were censused twice per month, between April and June, in 1996 and 1997 using the CBC methodology (Marchant et al. 1990). This method is well suited to monitoring yellowhammers, as they are sedentary, conspicuous and mono-territorial during the breeding season. Censuses were performed between 07:00 and 13:00 GMT, but were not carried out in wet or windy (> Force 4 on the Beaufort scale) weather. The route walked was reversed between censuses, to control for diurnal variation in yellowhammer activity. The locations of all territorial males and pairs were recorded and records from all censuses within a year were collated. Clustering of records was used to resolve the approximate position of territories. As it is impossible to define territory boundaries exactly, any inferences about territory size based on the outlines drawn around clusters are unreliable (Osborne 1984). Therefore, the number of territories on a boundary section (after controlling for section length) was used as the response variable in habitat selection models. A subset of four of the sites was censused more frequently (every 3 days) from April to July, in 1994 and 1995.

Nest recording

Once territorial pairs had been located, nests were found by systematic searching of the focal area of each pair's activity. Nests were discovered at various developmental stages, including nest building, incubation and brood rearing. Nests were visited at 3-day intervals to gain accurate data on all phases of the breeding cycle, recording the following data: first egg date, clutch size, date of hatching, number of nestlings hatched and number fledged, and eventual success or failure to fledge young. For nests that failed, the date and, where possible, the cause of failure were recorded (e.g. starvation, predation, nest collapse). The date of failure (or fledging) was estimated as the mid-point between the date when the nest was last known to be active and the date on which it was found to have failed (or the fledglings to have left the nest). When nestling age was not known precisely from observation of hatching, it could be estimated by comparing the degree of feather development of the largest nestling with known-age broods. Age estimates allowed back-calculation of first egg date for the nest, assuming a 13-day incubation period and a clutch completion period of 2–3 days, depending on clutch size (Cramp & Perrins 1994).

Statistical analysis


To find which variables explained a significant proportion of variation in yellowhammer territory density, we modelled counts of numbers of yellowhammer territories per boundary section as a function of habitat variables that varied between boundary sections. This was done separately for each year. This was achieved using multiple log–linear regression (counts of number of territories per boundary section, assuming a Poisson error distribution and log link; Crawley 1993). The logarithm of boundary length was specified as an offset (i.e. constraining this variable to have a regression coefficient of one) to control for the linear relationship between boundary length and number of territories on a boundary. In each case, significance of a habitat variable was tested by the change in deviance (ΔD, equating to a likelihood ratio test) of the model on removal of that variable. Quadratic terms were included to test for the possible non-linear effects of the continuous variables.

In significance testing, the effect of each habitat variable was first tested in isolation in a univariate test. All variables that explained a significant proportion of variation in territory density at P ≤ 0·05 were then entered in a multiple regression model. Intercorrelation between habitat variables was determined by calculating Pearson, Cramer or Phi coefficients, as appropriate (Siegel & Castellan 1988). When the coefficient between any two variables exceeded 0·5, alternative models were built, excluding each in turn. A backward selection procedure (Crawley 1993) was used, with least significant variables being removed sequentially, until a minimum adequate model (MAM) was reached in which all variables were retained at P ≤ 0·05. Log–linear regression prohibits the calculation of meaningful R2 values, as the variance is proportional to the mean in these models. Instead, goodness-of-fit of the MAM was tested by examining the ratio of the residual deviance to the residual degrees of freedom. Values over two for this ratio indicate overdispersion, i.e. a model that is poorly fitting due either to non-Poisson distributed error or failure to include important predictor variables (Crawley 1993).

Step-wise selection can lead to inappropriate selection of predictors and the final model can vary according to the selection procedure chosen (James & McCulloch 1990). To combat this problem and to minimize type I errors, it was considered that only predictors appearing consistently in the MAM for different years reflected robust ecological associations (Petraitis, Dunham & Niewiarowski 1996). This method may inflate the risk of statistical type II errors, but the large sample sizes in this study should have provided enough power to reduce the risk.

Boundary sections are not spatially independent and so our analyses incorporated a degree of spatial autocorrelation that increased the probability of type I errors. This potential problem was dealt with explicitly by forcing a dummy variable ‘farm’ (nine levels, identifying the individual study sites) into the MAM. Predictors becoming non-significant once farm was introduced were more likely to be affected by spatial autocorrelation, and their effects were interpreted with this in mind. This technique helps to identify variables that are most likely to have been acting as surrogates for other, unmeasured, variables acting at the farm level (Green, Osborne & Sears 1994).

For the purposes of illustrating quantitatively the predicted effect of specific management changes on yellowhammer territory density, new log–linear models, which contained only the most robust habitat variables (i.e. those which consistently predicted territory density in different years and which remained in MAMs when the factor farm was included in MAMs), were constructed for each year. In a baseline model, all of these variables were set to 0, i.e. absent. For each year, a series of new models was then constructed. For each model, just one of the predictors was set to ‘non-0’ and the effect on territory density of addition of that feature was quantified.

To understand further the predictive power of these models, we conducted a logistic regression analysis (binomial error and logit link; Manel et al. 1999) to model yellowhammer territory presence or absence on a boundary as a function of the same ‘robust’ predictors, as defined above. The fitted probabilities of occurrence were then interpreted as predicted presence, predicted absence or ‘prediction uncertain’, using a series of thresholds that allocated an increasing proportion of intermediate values (centred on 0·5) to the prediction uncertain category. A two-way ‘classification table’ (Norusis/SPSS Inc. 1994) of observed presence and absence vs. predicted presence and absence was then constructed (Fielding & Bell 1997).

Nest site location and breeding success

Variation in nest location, with respect to nest height above the ground, was modelled as a function of date, using normal errors and an identity link (i.e. without transformation of the response variable, nest height).

To find which variables explained a significant proportion of variation in timing of nesting attempts and in nest success, we modelled first egg date, clutch size, proportion of nestlings hatched that fledged, daily nest failure rates at the egg and nestling stage, and proportion of nests predated as a function of year (four-level factor), farm (nine-level factor), farm type (two-level factor, organic vs. intensive), first egg date (covariate), nest site type (1 = hedge, 2 = ditch) and land use adjacent to the nest (two-level factors denoting presence or absence of crops; field margin width as a covariate), using generalized linear models. A quadratic term for first egg date was included to account for simple curvilinear seasonal effects. Nest failure rates were also modelled as a function of observer visit rate (mean number of days between visits to a nest). In all model building, ‘noise’ variables (year and farm) were retained in models whether or not they were significant, while testing the effect of the parameter in question.

First egg date and clutch size were modelled using normal errors and an identity link. Daily nest failure rates were modelled using logistic regression (binomial errors and a logit link), with nest fate (failure = 1, success to hatching or fledging = 0) as the binary response variable and the number of days that the nest was exposed to the risk of failure after it had been found (‘exposure days’) as the binomial denominator. This follows the method recommended by Aebischer (1999). This method corrects for differences in the stage at which nests were found (assuming 3 days for completion of the average clutch, 13 days incubation, and 12 days from hatching to fledging; Cramp & Perrins 1994). An assumption of this method is that daily nest survival probabilities remain constant over the period for which an overall survival rate is calculated. For these analyses, the ‘egg stage’ was considered to span the period from laying of the first egg to hatching, and the ‘nestling stage’ to span the period from hatching to fledging. Similarly, variation in nest predation rates was analysed, using predation (predated = 1, success or other fate = 0) as the binary response variable. Whole brood starvation was also modelled by binary logistic regression (starved = 1, other fate = 0). Variation in the proportion of nestlings hatched that fledged was modelled using logistic regression (binomial errors and a logit link), with the number of nestlings fledged as the response variable and the number of nestlings hatched as the binomial denominator.

Statistical analyses were undertaken in glim release 4 (NAG 1993) and minitab release 12 (Minitab 1998). Mean values are presented with their standard errors, unless otherwise stated.


Settlement patterns

Territory densities at the farm scale varied between 0·07 and 0·29 territories per hectare, probably largely reflecting variation in hedgerow densities (Table 1).

Table 3 presents those predictors that appeared in over 50% of MAMs for each time period under consideration. Yellowhammer territory density (territories 100 m−1) was significantly higher on field boundaries with hedges and/or ditches than on boundaries that consisted of tree lines (often overgrown hedgerows) or isolated bushes (which are often relict hedgerows; Barr et al. 1992) in all 4 years. Density also increased with the width of any uncultivated grassy margin alongside boundary sections in all years and was higher on boundaries adjacent to rotational set-aside, relative to those adjacent to other field types, in 2 years. In contrast, boundaries with adjacent silage leys and/or grazed pastures supported lower territory densities than other boundaries in all 4 years. In univariate models, boundaries of fields under organic management held higher numbers of territories per section than those on intensive sites in 2 years. This effect was reversed in one of the other years. The effect of farming regime (organic vs. conventional) was only retained in one multivariate model. Positive associations between territory densities and oilseed rape and negative associations with roads, beans and gardens in univariate models were usually dropped when building MAMs. There was an inconsistent relationship between number of yellowhammer territories per section and adjacency of a farmyard. The ratio of residual deviance to residual degrees of freedom varied from 0·85 to 1·43, suggesting that the MAMs were well-fitting models.

Table 3.  Significant predictors of yellowhammer territory density. Predictors retained at P < 0·05 in the minimum adequate model (MAM) for each year are presented. + or – refers to the direction of the relationship between the predictor and yellowhammer territory density. * denotes that the predictor remained significant at P < 0·05 on inclusion of the factor ‘farm’ in the MAM. ‘Hedge’ and ‘Tree-line’ were strongly intercorrelated, and so alternative MAMs were developed, containing ‘hedge’ or ‘tree-line’ in turn. Predictors that significantly influenced density in either alternative are included in the table. + for farming regime in 1996 signifies that, in the MAM, organic farms held higher densities than intensive farms
Hedge presence**+ 
Tree-line presence   
Ditch presence  **
Road adjacent  
Width of margin*+*+
Pasture adjacent** *
Silage ley adjacent *
Set-aside adjacent+ * 
Winter rape adjacent   +
Beans adjacent   
Farmyard adjacent*  +
Gardens adjacent*   
Farming regime  + 
Residual deviance/ residual d.f.1·001·291·430·85

New models were constructed for each year, containing just the six ‘most robust’ predictors (width of grass margin, ditch presence, hedge presence, pasture adjacent, silage ley adjacent, rotational set-aside adjacent). Here the baseline model, in which values for all these variables were set at 0, represented in practice a line of isolated bushes or a tree line (the two extreme outcomes of over-intensive management of hedgerows or neglect) in an arable landscape, with neither a ditch nor a field margin. The effect on territory density of addition of each feature in each year is shown in Fig. 1.

Figure 1.

The effect on yellowhammer territory density (number of pairs per 100 m) of addition of each of six features to a ‘standard poor boundary’. In practice, at the sites studied, this represents a line of trees (or, more rarely, a line of isolated bushes) between arable fields, with no ditch and no field margin. (a) Addition of a ditch; (b) changing the boundary to a hedge; (c) placing a pasture field adjacent to the boundary; (d) placing a grass ley adjacent to the boundary; (e) placing a set-aside stubble next to the boundary; (f) adding a grass field margin of width 1 m to 10 m. In all cases, relationships are given for each of the 4 years of the study. * indicates that the parameter had a significant influence on yellowhammer territory density in the minimum adequate model for that year. ** indicates that the significant effect on yellowhammer territory density was retained when the factor ‘farm’ was added to the minimum adequate model. 95% confidence intervals are given (a–e). Confidence intervals (u, upper; l, lower) are omitted from (f) for clarity; these are, after back-transformation (1994: ucl + 0·029, lcl − 0·026; 1995: ucl + 0·028, lcl − 0·027; 1996: ucl + 0·014, lcl − 0·013; 1997: ucl + 0·016, lcl − 0·016).

The proportion of boundary sections on which yellowhammer territory presence or absence was correctly predicted by the logistic regression models is presented in the classification table (Table 4). In general, the models had high predictive power, but in all years presence (mean across all years and all cut-offs = 95%) was predicted very accurately and better than absence (mean across all years and cut-offs = 35%). In other words, the main weakness of the model was to predict that territorial yellowhammers should be present where, in fact, they were absent.

Table 4.  Classification tables for each year, generated from logistic regression models (yellowhammer territory presence or absence on a boundary), with predictors hedge, ditch, margin width, pasture, grass ley and set-aside. Boundaries were either classified as having a yellowhammer territory present or absent. PA, predicted absence; PP, predicted presence; OA, observed absence; OP, observed presence. Fitted probability varied from 0 to 1 and this was used to predict presence (PP) or absence (PA) using a series of four increasingly stringent cut-offs to exclude from consideration boundaries with intermediate values. Thus 0–0·49 = PA, 0·50–1·00 = PP (least stringent) to 0–0·19 = PA, 0·80–1·00 = PP (most stringent). The percentage of observed absences (OA–PA) and presences (OP–PP) that were correctly classified by the model are given for each year and cut-off
Cut-offs 0–0·49 : 0·50–10–0·39 : 0·60–10–0·29 : 0·70–10–0·19 : 0·80–1
OA-PA43·9 51·3 48·0 42·9 
OP-PP88·8 91·8 92·5 98·6 
OA-PA29·5 32·4 21·2 18·2 
OP-PP94·9 96·7 100 100 
OA-PA47·7 36·7 33·3 23·5 
OP-PP88·6 95·6 98·0 100 
OA-PA43·0 38·4 26·1 25·0 
OP-PP86·8 92·6 98·0 97·9 

Nest site location

Locations of discovered nests may reflect some observer bias. Nonetheless, almost all (99%) nests were either in herbaceous vegetation on ditch banks or in hedges. Of these, 43% of nests were found on organic farms. Nest height increased through the season (April 0·00 ± 0·00 m, May 0·23 ± 0·03 m, June 0·50 ± 0·03 m, July 0·71 ± 0·04 m, August 1·30 ± 0·12 m; F4,442 = 39·23, P < 0·001). This reflected a seasonal shift in nesting location from ditches to hedges (Fig. 2), although during the period when most nesting attempts took place, vegetated ditches remained the favoured location.

Figure 2.

Timing of initiation of yellowhammer clutches placed in herbaceous vegetation in ditches and in shrubby vegetation in hedges.

Breeding phenology and success

Timing of breeding

Earliest clutches were initiated on 6 May 1994, 28 April 1995 and 1996, and 29 April 1997 (Fig. 3). When the effects of annual and spatial variation were controlled for (by incorporating the factors year and farm), mean first egg date on organic farms was slightly but significantly earlier than on intensive farms (F1,465 = 3·970, P = 0·047, R2 = 2·1%). To reduce the influence of second clutches on mean first egg dates, these analyses were repeated for the first 30% of nests only. The significance of the organic/intensive contrast was increased, although still weak (F1,155 = 7·490, P = 0·007, R2 = 5·1%), reflecting a greater number of nests initiated on organic farms in the first week of May (Fig. 3). At the frequently censused subset of sites, individual yellowhammers were identified by colour rings and individual plumage variation, and thus the timing of successive breeding attempts of individual pairs could be determined (Kyrkos 1997). Of the 71 definite second attempts, 45 (63%) were replacements of failed first attempts, while only 26 (37%) followed a successful first attempt. A third clutch never followed two previous successful attempts (n = 8), suggesting that a maximum of two successful broods was raised, with most females making no more than three nesting attempts.

Figure 3.

First egg date distribution of yellowhammers nesting on organic and intensively managed farms. Note the tendency for earlier clutch initiation on organic farms in early May (day 30–40).

Clutch size

Clutch size varied from two to five eggs, with a median of three and a mean of 3·27 ± 0·03 (n = 442). Compared with other UK studies with similarly large samples (n > 400), the mean clutch size in this study was significantly smaller (Parkhurst & Lack 1946, 3·29, n = 809; O'Connor 1980, 3·50, n = 466; Crick, Donald & Greenwood 1991, 3·42, n = 1388; Yom-Tov 1992, 3·49, n = 809; Siriwardena et al. 2000, 3·42–3·59, n = 1028). However, samples in these studies overlapped considerably due to use of the nest record data set held by the British Trust for Ornithology (BTO). When annual and spatial variation were controlled, clutch size was found to vary seasonally, showing a quadratic relationship with first egg date of the form:

clutch size = 2·534  +  (0·022107  × date) - (0·000153  × date2).

This predicted a maximum mean clutch size of 3·33 on day 72 (11 June, first egg date: F1,396 = 6·34, P = 0·012; first egg date2: F1,396 = 6·92, P = 0·009). However, the variation explained by this effect was very small (R2 = 1·7%). When annual, seasonal and spatial variation were controlled, clutch size was not significantly influenced by farming regime (F1,396 = 1·04, P = 0·308), nest location or any land use next to the nest.

Breeding success

When annual and spatial variation were controlled, the proportion of nestlings fledged increased during the season (ΔD1 = 6·899, P < 0·01 n = 307). A mean of 2·60 ± 0·06 nestlings fledged per successful nesting attempt (n = 253). When annual and spatial variation were controlled, there was no seasonal trend in daily nest survival rates (first egg date: ΔD1 = 2·777, NS; first egg date2: ΔD1 = 0·284, NS). Neither farming regime nor any specific land use adjacent to the nest significantly influenced daily nest survival rate during either the egg or the nestling stage. Nest survival rates for each developmental stage are given in Table 5 and compared with those presented by Siriwardena et al. (2000) from analyses of nest record cards held by the BTO.

Table 5.  Survival rates of yellowhammer nests at different developmental stages on organic and intensively managed farms in Oxfordshire. Also included are the ranges in comparative rates calculated from nest record card (NRC) data held by the BTO (Siriwardena et al. 2000). Numbers in parentheses are variances
  Egg stageNestling stageFirst egg to fledge
OrganicDaily survival rate0·9700 (2·838 × 10−5)0·9715 (2·566 × 10−5) 
Nest success rate0·6730 (2·309 × 10−3)0·7068 (1·956 × 10−3)0·4757 (2·580 × 10−5)
IntensiveDaily survival rate0·9774 (1·917 × 10−5)0·9580 (3·119 × 10−5) 
Nest success rate0·7429 (1·871 × 10−3)0·5976 (1·748 × 10−3)0·4440 (2·406 × 10−5)
NRCDaily survival rate0·94–0·980·93–0·98 
Nest success rate0·40–0·680·44–0·780·18–0·52

Most nest failures were caused by predators (63·6%) at the egg (n = 84; 28·9% of failures) or nestling (n = 101; 34·7%) stages, although predator identity could not be determined reliably. Daily nest predation rates at the egg and nestling stage were not influenced by lay date, farm type, adjacent land use or distance of the nest to the nearest woodland edge or farm building. However, nestling predation rates were higher when nests were placed in a ditch than when placed in a hedge (vs. successful nests: ΔD1 = 4·057, P < 0·05, n = 342; vs. all other nests: ΔD1 = 4·481, P < 0·05, n = 368).

Abandonments accounted for 62 (21·3%) failures, and whole-brood starvation occurred in 17 (5·8%) failures. The remaining failures (9·3%) had a variety of causes, including destruction by agricultural activity (disturbance by cattle, hedge cutting and ditch maintenance) and clutch infertility. Whole-brood starvation was not associated with the presence of any particular crop type adjacent to the nest or to farming regime. The time between observer visits to nests that ultimately failed was greater than that between visits to nests that fledged young (ΔD1 = 12·097, P < 0·001), so we concluded that observer visits were not causing increases in nest failure rate in this study.

Demographic predictions

A simple demographic model following the approach recommended by Ricklefs (1973) allowed comparison of nest success with survival rate data in order to estimate whether breeding productivity was likely to be sufficient for population stability. Yellowhammer survival rates have been estimated from analyses of national ring recoveries held by the BTO (Siriwardena, Baillie & Wilson 1998). Prior to 1988, annual survival rate estimates were relatively constant, with means of 0·535 ± 0·034 for adults and 0·556 ± 0·074 for first year birds during periods of stable population (1966–77 and 1982–87). During periods of population increase (1962–65 and 1978–81), the corresponding figures were 0·557 ± 0·056 (adults) and 0·522 ± 0·104 (first year birds). From 1988 to 1994 survival rate estimates fell to 0·449 ± 0·102 for adults and 0·440 ± 0·132 for first year birds as the population declined, although the sample size of recoveries available limited analytical power so this change was not statistically significant (Siriwardena, Baillie & Wilson 1998).

Most first-year mortality occurs in the first few weeks after fledging (e.g. Anders et al. 1997; I. Hill, unpublished data), and analyses of BTO data were based on birds ringed when fledged and independent of their parents. These analyses therefore largely excluded the influence of mortality between fledging and independence, which should be regarded as unknown. However, over the range of these survival rate estimates and our own breeding success data, we could then predict the survival rate between fledging and independence (henceforth ‘post-fledging survival rate’) required for population stability. By reference to known post-fledging survival rates of other passerines, we could infer the population trend that would be expected to result from a given combination of breeding success and post-independence survival rates.

Productivity per pair for our study population is detailed in Table 6, with cumulative errors calculated according to Crick & Baillie (1996). In accordance with the findings of Kyrkos (1997) from his study of the population on one of our study sites, it was assumed that all yellowhammer pairs made a minimum of two and a maximum of three nesting attempts, with the third occurring only if one of the previous two ended in failure. The proportions of the population occupying each of the seven resulting categories of nest success (the extremes being SUCCESS–SUCCESS–STOP and FAIL–FAIL–FAIL–STOP) were then calculated from the Mayfield estimates of overall nest success rate from our data. This gave an estimated mean of 3·27 ± 0·07 nestlings fledged per breeding pair over the whole population (Table 6).

Table 6.  Predicted number of yellowhammer nestlings fledged by a closed population on intensive farms
Nesting historyProportion of population with each nesting history (determined using nest success rate)Number of chicks produced per pair  
Success–success0·444 × 0·444= 0·1972·63 × 2= 5·26
Success–fail–success0·444 × 0·556 × 0·444= 0·1102·63 × 2= 5·26
Success–fail–fail0·444 × 0·556 × 0·556= 0·1372·63 × 1= 2·63
Fail–success–success0·556 × 0·444 × 0·444= 0·1102·63 × 2= 5·26
Fail–success–fail0·556 × 0·444 × 0·556= 0·1372·63 × 1= 2·63
Fail–fail–success0·556 × 0·556 × 0·444= 0·1372·63 × 1= 2·63
Fail–fail–fail0·556 × 0·556 × 0·556= 0·1722·63 × 0= 0
Mean over whole population  = 3·27 ± 0·07 

Figure 4 shows the proportion of fledged young required to survive to independence (a maximum of 14 days in yellowhammers; Cramp & Perrins 1994) in order for a closed population to be self-sustaining, given a wide possible range of variation in mean number of young fledged per breeding pair (2·0–5·0) and annual survival rate of full-grown independent birds (0·4–0·6). For the range of breeding productivity observed in this study (approximately 3·0–3·5 young fledged per breeding pair), post-fledging survival rates of 0·45–0·55 (0·944–0·958 per day for the first 14 days post-fledging) would be required, given a post-independence annual survival rate of 0·55 (i.e. approximately equivalent to the best estimate for the period before the recent population decline). These would rise to 0·75–0·85 (0·980–0·988 per day for the first 14 days post-fledging) given a post-independence annual survival rate of 0·45 (i.e. approximately equivalent to the best estimate for the recent period of population decline).

Figure 4.

Survival rate of yellowhammers between fledging and independence that would be required for a closed population to remain stable given varying breeding productivity (young fledged per pair per year) and varying annual survival rates of independent birds. The line parallel with the x-axis represents the maximum post-fledging survival rate observed in other species (0·43; starling).


Settlement patterns

The territory density observed at the farm scale encompasses the mean density found by Kyrkos, Wilson & Fuller (1998) of yellowhammers breeding on CBC plots (0·134 ± 0·019 in 1988, and 0·105 ± 0·015 in 1993). A number of studies have examined the association between characteristics of the field boundary and aspects of the farmland bird community (Osborne 1984; Green, Osborne & Sears 1994; Macdonald & Johnson 1995; Parish, Lakhani & Sparks 1995; Stoate 1999). In this study, analysis of territory distribution patterns in consecutive years allowed us to detect habitat associations that are robust to the effects of stochastic annual variation. This quasi-experimental approach can be crucial to revealing true bird–habitat interactions, in the absence of defined manipulative experiments (Block & Brennan 1993).

The results of the logistic regression analyses show that the models constructed had strong predictive power and give confidence that the features discussed below had strong associations with presence or absence of yellowhammer territories. The greater tendency of the models to predict presence where yellowhammers are absent than vice versa suggests that the habitat factors examined do not currently limit populations despite losses of field boundary habitat in recent decades (Barr et al. 1992). This might suggest that another factor is presently limiting yellowhammer populations on lowland farmland and has caused the recent population declines.

Boundary features

The consistently higher numbers of territories per boundary section in hedges than in tree lines or lines of isolated bushes is not surprising, given that we found that the shrubby vegetation of hedges was a favoured nest site of the species, and concurred with the results of other studies (Green, Osborne & Sears 1994; Macdonald & Johnson 1995; Parish, Lakhani & Sparks 1995; Kyrkos, Wilson & Fuller 1998). The selection of field boundaries with ditches, a favoured nest site in this study and others (Stoate, Moreby & Szczur 1998), was also unsurprising. With the seasonal switch from ditch to hedge nest sites, these results suggest that the combination of hedgerow and vegetated ditch may be important in allowing some yellowhammer pairs to make multiple nesting attempts in one breeding season.

Features adjacent to boundaries

The width of the uncultivated grassy field margin had the strongest, most consistent, positive association with the number of territories per boundary section. Although not used for nesting in this study, these strips could be important because of enhanced nest concealment afforded by the strip's vegetation (Stoate, Moreby & Szczur 1998), although our analyses did not show a relationship between adjacent land use and nest success. Alternatively, these strips could be important because they support high densities of invertebrate food (Dennis, Thomas & Sotherton 1994). Yellowhammers have been shown to prefer to forage in uncultivated field margins when provisioning nestlings (Kyrkos 1997; Stoate, Moreby & Szczur 1998; Perkins et al. 1999). The use of these uncultivated unsprayed strips by yellowhammers echoes positive responses of other declining species to similar low-intensity management, e.g. skylarks Alauda arvensis L. and fields under organic management (Wilson et al. 1997), grey partridges Perdix perdix L. and reduced pesticide inputs in crops (Potts 1991) and cirl buntings Emberiza cirlus L. and rough pasture (Evans 1997).

Strong, consistent, negative associations of settlement patterns with pasture and silage leys in this study accord with those found in a larger-scale analysis of habitat associations of yellowhammers on CBC plots (Kyrkos, Wilson & Fuller 1998). In the past, unimproved grassland was selected by yellowhammers (Wild 1938; Williamson 1968). More recently, contradictory associations with improved grassland have been reported, with a similar number of positive (Parish, Lakhani & Sparks 1995) and negative (Green, Osborne & Sears 1994) associations reported in the literature. Intensive grazing and agricultural improvement of grasslands by reseeding, drainage and fertilization are known to reduce both floral and invertebrate diversity within swards (Haggar & Peel 1994). This may particularly reduce the abundance of invertebrate groups such as lepidopteran larvae, grasshoppers (Orthoptera), ants (Hymenoptera; Formicidae) and spiders (Araneae), which are important in the nestling diet of yellowhammers (Wilson, Arroyo & Clark 1996; Stoate, Moreby & Szczur 1998). Loss of invertebrates from improved grass fields has already been identified as one cause of the population decline in another British bunting, the cirl bunting (Evans 1997). In addition, the dense sward structures characteristic of improved agricultural grasslands may reduce birds' access to the invertebrates that are present (Odderskaer et al. 1997; Perkins et al. 2000).

Other land use

The positive association with set-aside stubbles might reflect the fact that these are favoured winter habitats of yellowhammers and many other farmland bird species that seek grain and weed seed (Wilson, Taylor & Muirhead 1996). Cirl buntings breed most successfully when winter food resources, in the form of stubbles, are in close proximity to summer resources (Evans 1997). It may be that similar requirements for summer and winter resources to be in close proximity have driven the observed yellowhammer territory settlement patterns. In the summer, however, yellowhammers show no greater selection of stubbles than expected relative to their availability, when foraging for invertebrates for nestlings (Kyrkos 1997; A.J. Morris et al., unpublished data).

Farming regime

The inconsistency of direction of association with farming regime and the general failure of farming regime to be retained in MAMs, suggest that farming intensity was not an important factor determining yellowhammer settlement in this study. Recent work suggests that many species are more abundant on organic than intensive farms (Chamberlain, Wilson & Fuller 1999), and a Danish study (Petersen 1994) has shown that yellowhammer densities are over twice as high on organic farms as intensive farms. However, other studies have shown no response of this species to farming intensity (Green, Osborne & Sears 1994; Christensen, Jacobsen & Nøhr 1996). The highest densities of yellowhammers in Britain still occur in the south-east of the country, in which intensive arable production dominates (Gibbons, Reid & Chapman 1993), further suggesting that they may be able to tolerate intensive arable agriculture better than other species.

Breeding success

Breeding phenology

First clutches were initiated slightly earlier on organic than on intensive farms. This may be explained by food supplies, which other studies have shown are often more abundant on organic farms (reviewed by Gardner & Brown 1998). Increased food supply in early spring is known to lead to faster gonadal growth and accumulation of reserves necessary for egg laying in females (Wingfield & Farner 1980). If organic farms do provide a greater winter food supply for birds, a response in first egg date is therefore possible. Timing of first laying can, in turn, affect the number of broods that a pair can raise in a season. A delay in nest initiation on intensive farms may reduce the number of broods that can be raised in a season, with important implications for the ability of a population to sustain itself. Indeed, while present on the organic farms, the double peak characteristic of some multi-brooded species was lacking on the intensive farms (Fig. 3). This also might suggest a more limited number of breeding attempts on intensive farms.

Nest success and reproductive output

The slightly smaller clutch size observed in this study than in studies based on nest record cards might indicate a limiting effect of food resources on breeding at present. Alternatively, given the curvilinear relationship between clutch size and lay date, this result might suggest a failure of nest record cards to include extreme date nests, which reduce the mean clutch size of the population in our study. The finding that most nesting failure was due to predation concurs with previous analyses of nest record cards (Crick et al. 1994). Use of plasticine eggs in artificial clutches containing zebra finch Taeniopygia spp. eggs during a concurrent experiment (Kyrkos 1997) suggested that most predation was by corvids (Corvus spp. or magpie Pica pica L.) or mustelids, with the remainder attributable to predators such as rodents, foxes Vulpes vulpes L. and grass snakes Natrix natrix L. The greater predation rate of nests in ditches (which was not linked to season) compared with the finding of Stoate, Moreby & Szczur (1998) that nest predation rates were greater among nests in hedges than those in wide field margins, is not readily explained. One possibility is that nest predators with different search tactics are present at different densities at the sites of the two studies.

The weak trend for a higher proportion of nestlings to be fledged later in the season parallels the seasonal increase in nest success in British cirl buntings, which has been attributed chiefly to increased availability of orthopterans as food for nestlings later in the season (Evans et al. 1997).

Demographic predictions

Combining our productivity data and recent survival rate estimates of fully grown birds (Siriwardena et al. 1998) suggests that, at present, a daily post-fledging survival rate of 0·980–0·988 is required over the first 14 days. There are no published estimates of survival rates of fledgling yellowhammers up to independence. There are few other passerines for which post-fledging survival rates have been reported, and the periods over which survival has been estimated are variable (daily survival rates 0·954–0·983: Table 7). If it is assumed that mortality peaks just after fledging and that yellowhammer post-fledging survival rates are within the range reported for these other species, then these figures suggest that yellowhammer population stability in Oxfordshire is at present unlikely.

Table 7.  Passerine post-fledging survival rates reported in the literature
SpeciesAuthor(s)Days from fledging over which survival rate wasEquivalent constant daily survival rate measuredSurvival rate over period of measurement
Great tit Parus major L.Dhondt (1979)700·9810·261
Yellow-eyed junco Junco phaeonotusSullivan (1989)220·9540·355
Zebra finch Taenopygia guttataZann & Runciman (1994)350·9690·332
European starling Sturnus vulgaris L.Krementz, Nichols & Hines (1989)490·9830·432
Wood thrush Hylocichla mustelinaAnders et al. (1997)210·9600·424
Song thrush Turdus philomelosI. Hill, unpublished data350·9650·287
Blackbird Turdus merulaI. Hill, unpublished data350·9670·309

In accordance with this hypothesis, analyses of trends in breeding performance of yellowhammers at the national scale, using BTO nest record data (Siriwardena et al. 2000), showed that nest success rate and brood sizes have been either stable or increasing since the early 1960s. Data from our study show that nest survival rates at both the egg and nestling stage are at the high end of the range recorded (Table 5). This suggests that there has been no decline in the productivity of individual nests that could account for the population decline. Indeed, there may have been a density-dependent response to lower breeding densities. However, the number of nesting attempts made per pair are not documented in the historical data, so a reduction in this component of breeding productivity cannot be ruled out as contributing to the population decline. This could act through a shortening of the breeding season, an increase in the interval between nesting attempts or, just possibly, an increase in the proportion of the population that is unable to breed at all.

The demographic mechanisms underlying the recent decline of yellowhammers in regions covered by the CBC and the longer-term declines in western areas of the British Isles remain uncertain. However, while the number of nesting attempts and the proportion of the population that can breed are potentially limiting, it seems likely that the possible fall in survival rates recorded by Siriwardena et al. (1998) is presently the principal factor driving the decline in the yellowhammer CBC index since the late 1980s. This is compatible with the logistic regression equation in our study, which failed to implicate a breeding factor. Yellowhammers and other granivorous passerines are known to prefer grain and weed-rich foraging habitats such as stubble fields and grain spills outside the breeding season (Donald & Evans 1994; Evans & Smith 1994; Wilson, Taylor & Muirhead 1996). Weed-rich stubbles are rarer and bird-proofing of grain stores more common in the farmland environment than in the past (Evans 1997). It is possible that, in the absence of traditional winter food sources, yellowhammers could be more dependent on the use of gamebird feed sites and game-cover strips (Stoate & Szczur 1997), as with corn buntings Miliaria calandra L. and other granivorous passerines (Brickle 1997). These observations hint at the possibility that early spring (when seed availability is likely to be at its minimum, invertebrate activity remains low, and birds must attain breeding condition) may be a time of year when agricultural intensification has led either to increased mortality rates or a reduction in the ability of birds to attain good breeding condition.

Conclusions and conservation implications

If survival of fledged birds is presently limiting yellowhammer populations on lowland farmland, then breeding densities may benefit from an increase in the availability of seed resources. A variety of mechanisms could deliver seed-rich habitats, including spring-sowing of cereals with the retention of stubbles until late spring, game-cover strips and the managed return of harvest waste (under-sized grain and weed seed) to the land.

The results of this study suggest that ideal nesting habitat for breeding yellowhammers on lowland farmland in southern England consists of hedgerows associated with vegetated ditches and wide, uncultivated, grassy margins. Removal of hedgerows, filling or clearing of ditches, abandonment of hedge management allowing hedges to develop gradually into lines of trees with minimal shrub structure, and cropping or grazing right up to the field edge, all reflect removal of suitable nesting habitat for yellowhammers. UK initiatives such as ‘beetle banks’ (grass strips in cereal fields), game-cover strips and grass set-aside field margins would be expected to increase food and nest site availability for yellowhammers and other declining species, such as skylark (Wilson, in press), corn bunting (Brickle et al. 2000) and cirl bunting (Evans 1997).

Hedgerows adjacent to improved pastures and silage leys tend to be avoided by territorial yellowhammers. Given the fact that high densities of breeding yellowhammers may still be found on areas of semi-natural or unimproved agricultural grassland such as downland and heathland (Fuller 1982), this suggests that loss of invertebrate and/or seed food supplies associated with intensification of grassland management (Fuller 1987; Haggar & Peel 1994) may reduce breeding yellowhammer densities. It is noteworthy that northern and western regions of the British Isles have seen the greatest yellowhammer declines (Gibbons, Reid & Chapman 1993; Donaghy 1998). These are the areas that have become increasingly specialized towards intensive pastoral agriculture over recent decades whilst also losing the grain and arable weed seed food resources associated with arable production (Lack 1992). This echoes the earlier loss of breeding corn buntings from the same areas (Donald 1997). Research is needed to determine whether breeding densities of buntings would respond to extensification of agricultural grassland management, for example via lower stocking densities or the setting aside of ungrazed or unharvested margins around pastures and silage leys.

Retention of non-cropped habitat, extensification of grassland management and retention of winter stubbles are not economically competitive practices today. To restore breeding populations of yellowhammers and other species on lowland farmland, such practices will need to be favoured by policy reforms that divert subsidy support for production goals to biodiversity enhancement. The Arable Stewardship Pilot Scheme introduced by the Ministry of Agriculture, Fisheries and Food in England (MAFF) in 1998, in delivering grass margins, beetle banks, wildlife seed mixtures in fields and over-wintered stubbles, could help populations of yellowhammers and other declining species to recover in arable and mixed farming areas, if it became available nationally. At present, it is not obvious how population recovery can be assisted in pastoral areas, except by encouraging low-intensity small-scale cropping to provide on-farm winter fodder (Bignal & McCracken 1996).


We thank Helen Browning, Phil Douthwaite, Richard Green, Richard Manors, Pat and Daphne Saunders, David Sharpe, Adam Twine, Alastair Welford, The National Trust and the University of Oxford for access to land owned or managed by them. This work was also made possible by the great assistance in the field of Elaine Irving, Darren Moorcroft, Nicholas Wilkinson and Phil Barnett. Earlier drafts of the manuscript were improved by comments from Paul Donald, Andy Evans, Ian Hill, Gavin Siriwardena, Chris Stoate and Mark Whittingham. Ian Hill and Gavin Siriwardena kindly allowed access to unpublished data. The work was supported by a research grant from the Biotechnology & Biological Sciences Research Council and by a studentship to Antonios Kyrkos from the Rhodes Trust.

Received 18 June 1999; revision received 14 April 2000