Long-term changes in over-winter survival rates explain the decline of reed buntings Emberiza schoeniclus in Britain


Will Peach, Royal Society for the Protection of Birds (RSPB), The Lodge, Sandy, Beds SG19 2 DL, UK (e-mail will.peach@rspb.org.uk).


1. The reed bunting Emberiza schoeniclus is one of a suite of granivorous farmland bird species that suffered a major population decline in Britain during the late 1970s and early 1980s. Extensive monitoring data indicate a large increase in the abundance of reed buntings between 1963 and 1975, followed by a decline of 58% on farmland and 66% along linear waterways during the period 1975–83. Since 1983 numbers have remained relatively stable in both habitats.

2. During the population decline, breeding numbers declined rapidly on arable and mixed farms, but remained relatively stable on pastoral farms. The decline on farmland was greater in northern Britain than in the south-east.

3. Extensive nest recording indicated that breeding performance was higher during the period of population decline (2·74 young per nesting attempt) than during the preceding period of population increase (2·65) or the recent period of population stability (2·17).

4. Minimum survival during the first year of life (estimated from mark–recapture data) declined during the 1970s and early 1980s, and increased strongly during the 1990s. The trend in first-year survival was independent of a weak positive relationship with winter temperature. Although there was evidence of a similar temporal trend in adult survival, this disappeared when winter temperature was taken into account.

5. A demographic model indicated that the timing and magnitude of the observed changes in first-year survival and adult survival were each sufficient alone to account for the observed changes in the abundance of reed buntings during the period 1969–87. A reduction in over-winter survival was probably the main demographic cause of the reed bunting population decline, although loss of breeding habitat and a recent reduction in breeding performance may also have influenced numbers.

6. During winter reed buntings feed mainly on small grass and weed seeds. The observed declines in abundance and survival rates coincide with the widespread introduction of a range of efficient herbicides and the loss of winter stubbles from British farmland. Our findings are therefore consistent with the hypothesis that the decline of the British reed bunting population was caused primarily by a reduction in food availability outside the breeding season. Changes in agricultural practices that increase the abundance of small weed and grass seed on farmland during winter are likely to allow at least a partial recovery of the British reed bunting population.


Since the early 1970s a wide range of farmland bird species have undergone substantial population declines, both in Britain and elsewhere in Europe (Marchant et al. 1990; Tucker & Heath 1994; Fuller et al. 1995). Intensive studies of four of the most severely affected species (grey partridge Perdix perdix, corncrake Crex crex, lapwing Vanellus vanellus and skylark Alauda arvensis) have shown that changes in agricultural practices have reduced breeding productivity and, in each case, this has probably been a major cause of the population decline (Potts 1986; Galbraith 1988; Stowe et al. 1993; Wilson et al. 1997). For most other species the causes of population declines have yet to be determined.

Prominent amongst the suite of declining farmland bird species are small seed-eating passerines. Six seed-eating passerines were included on the 1996 Birds of Conservation Concern ‘Red List’ (Gibbons et al. 1996) as a consequence of major population declines (skylark, tree sparrow Passer montanus, linnet Carduelis cannabina, reed bunting Emberiza schoeniclus and corn bunting Miliaria calandra) or range contraction (cirl bunting Emberiza cirlus) (Evans 1992). More recently, yellowhammer Emberiza citrinella numbers have also declined on British farmland (Crick et al. 1998; Siriwardena et al. 1998). Evidence for increasing nesting success amongst tree sparrows, corn buntings and yellowhammers (Crick 1997; Crick et al. 1998), coupled with evidence for a general reduction in seeds and seed-rich habitats on farmland (O’Connor & Shrubb 1986; Donald 1998) particularly during winter (Barr et al. 1993; Shrubb 1997), suggests that reduced over-winter survival may have been a major cause of decline amongst granivorous farmland birds. Until now, however, demographic evidence supporting or refuting this hypothesis has been lacking.

In this study we analysed long-term demographic data to describe the population dynamics of reed buntings Emberiza schoeniclus L. in lowland Britain since the early 1960s. The reed bunting is an ideal subject for such a study because extensive information on changes in abundance and nesting success is available for two important breeding habitats (farmland and linear waterways), and because changes in survival rates can be derived from mark–recapture data collected over three decades at widely dispersed study sites. We used these data to test the hypothesis that a reduction in over-winter survival was the primary demographic cause of the population decline.


Temporal trends in abundance

Changes in the abundance of breeding reed buntings were assessed through analysis of annual census data collected on lowland farmland since 1962 and along linear waterways (rivers and streams) since 1974. These data were collected by skilled volunteer census workers participating in the Common Birds Census (CBC) and Waterways Bird Survey (WBS) organized by the British Trust for Ornithology (BTO) (Marchant et al. 1990). Both surveys employ the territory mapping technique to assess the numbers of breeding pairs present on each census plot. The CBC has been the subject of a large number and wide range of validation studies (Marchant et al. 1990), including its suitability for the monitoring of reed buntings (Bell, Catchpole & Corbett 1968), and is generally considered to provide reliable extensive information on population changes for a wide range of lowland bird species. Since the mid-1960s approximately 100 farmland CBC plots have been censused each year, and 40–60 of these have typically held breeding reed buntings. The numbers of WBS plots censused each year has increased from approximately 50 in the mid-1970s to more than 100 during the 1990s. Of these, 30–60 plots have typically held breeding reed buntings. Farmland CBC plots typically cover 70–90 ha dominated by farmland but also encompassing wooded areas and human sites such as farmyards; WBS plots typically cover linear waterways that are 3–6 m wide and 4–5 km long (Marchant et al. 1990). Recent data indicate that approximately 62% of British reed buntings breed on farmland (50%) or the preferred wetland habitats (12%), the rest nesting in semi-natural grassland, heath and scrub (Gregory & Baillie 1998).

Census data were analysed using log-linear Poisson regression, a form of generalized linear model with Poisson error terms for count data (McCullagh & Nelder 1989). Initially we modelled the expected annual count at each plot as the product of year and plot effects with the assumption that between-year changes in abundance were homogeneous across plots. Year effects were taken as the best estimates (or indices) of changes in abundance over time. This approach was extended to test for differences in changes in abundance between farm types and between regions during the period 1970–95 (1974–95 for WBS data). CBC plots were classified into three farm type groups (arable, grazing and mixed), and CBC and WBS plots were classified into three regions (south-east, south-west and northern Britain) (see Appendix 1 for definitions). Lack of data from the south-west of Britain forced us to limit regional comparisons to the north and south-east. As a simplification of the initial plots-by-years model, we constrained annual changes in abundance to be linear (on a log scale) over time. The significance of the slope parameter provided a statistical test for an overall change in abundance during the study period, and a means of comparing rates of change in abundance across farm types and regions. Although actual trends in abundance are unlikely to have been linear, the linear change model provided a convenient means of comparing overall changes in abundance between different farm types and regions.

Temporal changes in nesting success

Information on changes in nesting success was obtained from BTO nest record cards (Crick & Baillie 1996). Nest record cards provide information on various components of nesting success, as well as on nest location and the surrounding habitat. Data were extracted on five nesting success parameters during the period 1962–95: clutch size (CS), brood size (BS), chick : egg ratio (CER) and daily nest failure rates at the egg (EFR) and nestling (NFR) stages. Clutch and brood sizes were defined as the maxima recorded. The chick : egg ratio was defined as the proportion of eggs hatching in nests where the whole nest did not fail, i.e. brood size/clutch size. Our measure of brood size is likely to overestimate the brood size at fledging, but will approach it if mortality early in nestling life is the most significant form of partial brood loss. The CER will therefore incorporate early losses as well as hatching success (the proportion of the eggs in the clutch that hatch successfully). Brood sizes of zero were omitted from the calculation of CER so that complete failures were incorporated only in EFR. The data for all five variables were analysed using generalized linear models. Daily failure rates were estimated using models with failure (1) or success (0) of a nest over a given exposure period as the dependent variable, and the number of days over which the nest was exposed to failure as the binomial denominator (Crawley 1993; R. E. Green, personal communication). This approach represents a generalization of the standard Mayfield method (Mayfield 1961, 1975). The models for CS and BS used identity link functions and a normal error distribution, while those for CER, EFR and NFR used logit links and binomial errors for modelling proportions.

Limited sample sizes (Appendix 2) prevented the estimation of precise annual region- and habitat-specific components of nesting success, so we combined the annual samples according to blocks of years determined according to significant turning points in the CBC and WBS index series (Siriwardena et al. 1998). Three time periods were identified, when abundance was increasing (1962–74), decreasing (1975–83) and stable (1984–95). Analyses were repeated using a shorter decline period (1976–78) reflecting the period during which the largest and most sustained reduction in abundance occurred both on farmland and along linear waterways. Time period, region and habitat were each considered as categorical variables. The regions used were the north, south-east and south-west of Britain (Appendix 1), and the habitats were ‘farmland’ and ‘waterside’ as described by observers using the habitat codes of Crick (1992). Those cards not readily assignable to farmland or waterside habitats were omitted from habitat-specific analyses.

The significance of time period, region and habitat effects was assessed using likelihood-ratio tests between nested models. Models allowing simultaneous temporal and regional or habitat-specific variation were also fitted (using the longer decline period), and the significance of the interaction term was assessed. All models were fitted using proc genmod of SAS (SAS Institute Inc. 1996).

We also estimated overall nesting success as the mean number of fledglings raised per breeding attempt for each time period, habitat and region. This was calculated as:

CS × CER × (1 – EFR)ED × (1 – NFR)ND

where ED and ND are the lengths of the egg and nestling stages in days. ED and ND were taken to be the mid-points of the ranges given in Cramp & Perrins (1994): 16 and 11 days, respectively (allowing for 3 days of egg laying prior to incubation in ED). Confidence intervals for the estimates of nesting success were calculated using an extrapolation of the formula provided by Hensler (1985).

To investigate interannual variation in nesting success, we estimated annual daily nest failure rates through the egg and nestling stages combined (i.e. from laying to fledging) for the period 1962–95. Temporal trends in this overall estimate of laying-to-fledging nest mortality were identified by treating year (and year2 and year3) as continuous covariates in the models.

Temporal changes in survival rates of fully-grown birds

Annual survival rates were estimated through application of modified Cormack–Jolly–Seber multinomial probability models (Lebreton et al. 1992) to capture–mark–recapture data collected during the summer (April–September inclusive) at four reed bed sites spread across southern and central England. These sites were Marsworth Reservoir, Hertfordshire (1969–97), Chew Valley Lakes, Avon (1971–97), Brandon Marsh, Warwickshire (1972–97) and Wicken Fen, Cambridgeshire (1969–89). The farmland surrounding Marsworth Reservoir, Brandon Marsh and Wicken Fen is predominantly arable, with smaller areas of permanent pasture (at Marsworth and Brandon) and sugarbeet and vegetables (at Wicken) constituting the other main land use. The farmland surrounding Chew Valley Lakes is predominantly permanent pasture with smaller areas of arable and ley grassland. While catching effort has remained approximately constant at each of the four study sites, catches of reed buntings have fallen in size in line with the national population declines (see below). As reed buntings can be reliably aged as either juveniles (i.e. fledged during current breeding season) or adults (at least 1 year old), the recapture data were first separated into birds first caught as juveniles and first caught as adults. Analyses covered the period 1969–97 (28 annual survival periods) and sample sizes are summarized in Appendix 2.

Prior to survival modelling, two key assumptions of multinomial mark–recapture models (independence of fates of individuals and independence of future and previous capture histories) were checked for all seven data sets using the goodness-of-fit tests provided in program release (Burnham et al. 1987; Lebreton et al. 1992). These tests indicated that these assumptions were satisfied for each of the seven data sets, although for birds first caught as juveniles there was strong evidence of lower return rates during the first year after capture. This probably reflects a combination of lower survival and higher dispersal rates amongst young birds, and highlighted the need for age-specific model parameters.

Initial survival modelling was carried out for each site separately. Models were fitted with year-specific survival rates and recapture probabilities for each of the two age classes of birds. Attempts were then made to simplify this starting model, first by constraining recapture probability to be constant across years, and then to be constant across age classes. For each study site there was no evidence that recapture probabilities varied significantly between years, but strong evidence that they differed between first-year and older birds. The lack of interannual variation in recapture probabilities probably reflects the approximately constant annual capture effort maintained at each site. Survival rates at individual sites were found to differ between age classes but not between years.

Subsequent analyses involved combining the recapture data from the four study sites into a single multisite analysis. By combining information in this way we improved the precision of our annual survival estimates and increased our power to detect temporal changes in survival. The starting model was one in which recapture probability varied between sites and age-classes but not between years (two parameters per site), and survival varied between years and between age-classes (adults and first-years) but not between sites (56 parameters). Attempts were made to simplify recapture and survival rates (by fitting models with fewer parameters). The analysis proceeded in three stages. First, we tested for interannual variation and long-term temporal trends in survival rates. Formal testing for long-term trends in survival is useful because such trends may not be obvious when year-specific survival estimates are imprecise and variable between years. Secondly, we checked that the observed trends in survival were not artefacts of confounding trends in recapture probability. Thirdly, we tested for relationships between annual survival and winter temperature because reed buntings can suffer high mortality during severe cold weather (Prys-Jones 1984). Our measure of winter temperature was the central England mean surface air temperature series (Manley 1974) averaged across the months of November–February. All survival models were fitted using software surge4 (Pradel, Clobert & Lebreton 1990). Akaike’s information criterion (AIC) was used as the main guide to model selection and likelihood-ratio tests were used to test specific hypotheses (Lebreton et al. 1992).

Modelling the implications of demographic changes on population levels

In order to test whether the observed changes in survival rates were large enough to have caused the observed changes in abundance, we constructed a time-varying Leslie matrix model (McDonald & Caswell 1993) in which breeding productivity (the numbers of fledged independent young reared per year per breeding territory) was held constant over time. All birds were assumed to breed when they were 1 year old (Cramp & Perrins 1994), and year-specific survival estimates (from recapture analyses) were used to predict abundance indices. When either first-year or adult survival was constrained to be constant, we used average estimates derived from national ring recovery data (0·542 for adult survival and 0·474 for first-year survival; Siriwardena, Baillie & Wilson 1998), which are less vulnerable to the underestimation biases that can affect mark–recapture estimates (caused by permanent emigration from study sites).

The productivity measure used in these models was estimated iteratively as the constant that minimized the squared deviations between the observed and predicted population indices. Although our mark–recapture survival estimates probably underestimated true survival (as a consequence of permanent emigration), the demographic model allowed for this by inflating the productivity constant to the level needed to balance gains and losses from the population. The model therefore assessed whether the timing and magnitude of changes in survival rates were likely to have been important determinants of changes in population levels. The model was not intended to assess whether the absolute rates of survival were likely to have caused changes in abundance.


Temporal changes in abundance

Changes in the abundance of reed buntings breeding on farmland and along linear waterways are shown in Fig. 1. Numbers on farmland increased steadily during the period 1963–75 following the severe winter of 1962–63 (Marchant et al. 1990). Abundance then declined rapidly between 1976 and 1979 (by 49%) and again in 1982 (Fig. 1), since when there has been a slow and relatively small increase in numbers. Overall abundance increased by 348% between the low point of 1963 and the peak of 1975, and declined by 54% between 1975 and 1995. A similar pattern of change was evident along linear waterways (Fig. 1), with numbers declining by 66% between 1975 and 1983 and remaining relatively stable thereafter. Numbers on farmland fell by 58% between 1975 and 1983. There was no significant difference between the average annual rates of decline (from linear change models) on farmland and linear waterways between 1974 and 1995 (Wald test: inline image = 0·02, P = 0·88).

Figure 1.

Annual indices of reed bunting abundance on (a) farmland and along (b) linear waterways. Indices (and 95% confidence limits) were derived using log-linear models and are plotted on a logarithmic scale.

Reed buntings declined significantly on both arable and mixed farms between 1970 and 1995 (by 6·5% and 4·6% per annum, respectively) but remained relatively stable on pastoral farms (average rate of change was +0·004% p.a.). These average rates of change differed significantly between farm types (Wald test: inline image = 36·8, P < 0·0001). Over the same period in northern Britain, reed buntings declined by 7·2% per annum on farmland and by 5·2% per annum along linear waterways. In south-eastern Britain, farmland counts declined by 4·1% p.a. and waterways counts declined by 3·8% p.a. The higher rates of decline in the north differed significantly from those in the south for farmland counts (Wald test: inline image = 12·0, P < 0·001) but not for waterways counts. The decline on northern farmland was larger and more prolonged than that on south-eastern farms (Fig. 2).

Figure 2.

Annual indices of reed bunting abundance on farmland in northern (squares) and south-eastern (triangles) Britain. Indices are derived using log-linear models on Common Birds Census data and regions are defined in . Appendix 1

Temporal changes in nesting success

There was no evidence that any components of nesting success were lower during the period of population decline (1975–83) (see Appendix 2 for statistical details). The chick : egg ratio was significantly lower, and egg stage nest failure rate significantly higher, during the period of population stability (1984–95) than during the earlier periods of increase and decline. Although average clutch size was smaller during the shorter decline period (1976–78) (Appendix 2), overall nesting success was still highest during the decline period and lowest during the period of population stability (Fig. 3a).

Figure 3.

Temporal variation in nesting success (a) during periods of population increase (1962–74), decrease (1975–83) and stability (1984–95), (b) in three regions of Britain, and (c) in farmland and wetland habitats. Error bars are 95% confidence limits.

There were significant regional differences in brood size and nestling stage failure rate (Appendix 2). Both variables indicated that nests in northern Britain were relatively successful, and this resulted in higher overall nesting success per breeding attempt in the north. Temporal variation in brood size and nestling stage failure rates also differed significantly between regions, such that nesting success in northern Britain declined across the three periods, while in south-east Britain nesting success was highest during the decline period (Fig. 3b). In all three regions overall nesting success was lowest during the period of population stability (1984–95). There were no significant overall differences among the components of nesting success between nests in farmland and waterside habitats, but significant interactions between habitat and time period for brood size, clutch size and chick : egg ratio indicated that nesting success underwent different patterns of temporal change in the two habitats (Appendix 2). In both habitats, nesting success was highest during the period of population decline and lowest during the period of stability (Fig. 3c).

These results suggest that changes in nesting success were probably not the cause of the decline in the reed bunting population during the late 1970s and early 1980s. Further support for this conclusion was provided by the analysis of interannual variation in laying-to-fledging daily nest failure rates. A significant amount of the interannual variation in overall nest failure rate was explained by a non-linear trend in which failure rates declined during the 1960s and early 1970s and then increased slowly during the 1980s (Fig. 4). There was no indication of relatively high nest failure rates during or immediately prior to the period of population decline (1976–83).

Figure 4.

Annual estimates (and 95% confidence limits) of egg-to-fledging daily failure rates estimated from all available nest record cards. The line shows the trend predicted from a cubic function of year. The significance of the cubic trend was assessed using likelihood-ratio tests ( inline image = 3·4, P = 0·065 vs. a quadratic trend; inline image = 16·5, P < 0·001 vs. a model with no temporal trend) and AIC [6495·1 (cubic) vs. 6496·5 (quadratic) vs. 6505·5 (linear) and 6505·7 (no trend)].

Temporal changes in survival rates

Survival rates of both adult and first-year reed buntings varied significantly between years (Table 1). Despite the poor precision of individual year-specific survival estimates, there was strong evidence for a non-linear trend in first-year survival (Table 1) in which survival declined during the late 1970s and 1980s, and then increased strongly during the 1990s (Fig. 5a). There was weaker evidence (P < 0·1) for a non-linear trend in adult survival (Fig. 5b). Although these trends do not account for much of the interannual variation in survival, they are useful in demonstrating a statistically significant decline in average survival rates at about the time the British breeding population was declining. Although the temporal trends in first-year and adult survival appear similar (Fig. 5), models with identical trends but different intercepts (models 12 and 13 in Table 1) were rejected, indicating that the temporal changes in survival differed between adult and first-year birds.

Table 1.  Modelling the survival rates of reed buntings during the period 1969/70–1996/97 using summer capture–mark–recapture data collected at four long-term study sites in England. Models are defined in terms of survival (Φ) and recapture rates (p) that can differ between first-year (fy) and adult (ad) birds. Parameters can be year-specific (t), site-specific (s) or constant (c). Temporal trends in survival and recapture rate were also modelled as cubic (cub), quadratic (quad) and linear (lin) functions of year (on a logit scale). In models 1–13 and 17–20 recapture rate differed between adults and first-years and between sites but was constant over time (eight parameters). The number of identifiable parameters (np), relative deviance (dev) and Akaike’s information criterion (AIC) values are presented for each model. Models with the smallest AIC values provide the most parsimonious descriptions of the data, and are therefore preferred. At each stage of the analysis the preferred model is highlighted in bold
Model npdevAIC
  1. Likelihood-ratio tests: Interpretation:

  2. Model 1 vs. 2: χ227 = 41·9, P < 0·05 Significant interannual variation in adult survival

  3. Model 1 vs. 3: χ227 = 62·0, P < 0·001 Significant interannual variation in first-year survival

  4. Model 5 vs. 3: χ23 = 23·0, P < 0·001 Significant cubic trend in first-year survival

  5. Model 8 vs. 10: χ22 = 4·9, P < 0·1 Weak evidence of quadratic trend in adult survival

  6. Model 10 vs. 12: χ22 = 10·5, P < 0·01 Adult survival and first-year survival do

  7. Model 10 vs. 13: χ23 = 11·9, P < 0·01 not share the same temporal trend

  8. Model 15 vs. 16: χ23 = 4·7, P > 0·1 No evidence for a confounding cubic trend in recapture probability

  9. Model 17 vs. 4: χ21 = 4·8, P < 0·05 Winter temperature explains significant interannual variation in first-year survival

  10. Model 18 vs. 17: χ23 = 25·0, P < 0·001 Cubic trend remains highly significant in the presence of the effect of winter temperature

  11. Model 18 vs. 8: χ21 = 1·5, P > 0·2 Effect of winter temperature not significant in the presence of the cubic trend

  12. Model 19 vs. 2: χ21 = 2·8, P < 0·1 Weak evidence that winter temperature explains interannual variation in adult survival

  13. Model 20 vs. 19: χ22 = 2·8, P > 0·2 No evidence for a quadratic trend in adult survival having allowed for the weak relationship with winter temperature

Testing for year-specific survival
Testing for trend in first-year survival
5Φ(fyt = cub,adt)406469·86549·8
6Φ(fyt = quad,adt)396474·26552·2
7Φ(fyt = lin,adt)386491·16567·1
8Φ(fyt = cub,adc)136516·26542·2
Testing for trend in adult survival
9Φ(fyt = cub,adt = cub)166511·36543·3
10Φ(fyt = cub,adt = quad)156511·36541·3
11Φ(fyt = cub,adt = lin)146515·26543·2
Testing for additive trends in first-year and adult survival
12Φ(fyt + adt = cub)136521·86547·8
13Φ(fyt + adt = quad)126523·26547·2
Checking for confounding trends in recapture probability
14Φ(fys,ads),p(fyt = cub,adc)136537·66563·6 (cf. model 8)
15Φ(fyt = cub,adc),p(fyt = cub,adc)106546·36566·3
16Φ(fyt = cub,adc),p(fyc,adc)76551·06565·0
Modelling the combined effects of winter temperature and trend
17Φ(fyt = wtemp,adc)116539·76561·7
18Φ(fyt = wtemp + cub,adc)146514·76542·7
19Φ(fyt,adt = wtemp)386469·96545·9
20Φ(fyt,adt = wtemp + quad)406467·16547·1
Figure 5.

Minimum annual survival probabilities (and 95% confidence limits) of (a) first-year and (b) adult reed buntings estimated from mark–recapture data. The solid curve shows the significant cubic trend fitted through the first-year survival probabilities, and the dashed curve shows the weakly significant quadratic trend in adult survival (see Table 1 for details).

To check that the apparent trend in first-year survival was not an artefact of a confounding trend in recapture probability, we fitted three urther models to the data. First, we reversed the survival and recapture components of model 8 to test whether the data were better described by a non-linear trend in recapture probability (model 14). The much higher AIC value for model 14 (Table 1) confirms that the non-linear trend lies in the survival component of the model and not in the recapture component. We then tested for a non-linear trend in recapture probability in the presence of a non-linear trend in survival (models 15 and 16 in Table 1), but found no evidence for the former.

Although there was a weak (r2 = 17%) positive relationship between first-year survival and winter temperature, the non-linear temporal trend remained highly significant after the variation associated with winter temperature had been controlled for (model 17 vs. 18; Table 1). There was also weak evidence for a relationship between adult survival and winter temperature (model 19 vs. 2; Table 1), and when this was controlled for there was no evidence of any trend in adult survival (model 19 vs. 20).

Implications of changes in survival rates

Leslie matrix models incorporating interannual variation in either first-year or adult survival were similarly successful (as measured by the sum of the squared deviations between the observed and predicted population indices) in predicting both increases and decreases in the abundance of reed buntings between 1969 and 1996 (Fig. 6). Although incorporating interannual variation in both first-year and adult survival rates increased the deviations between the observed and predicted population changes (Fig. 6), this model still predicted consecutive periods of increasing abundance (1969–76), steep decline (1976–82) and more gradual decline and stabilization (1982–96).

Figure 6.

Predicted changes in the abundance of reed buntings on (a) farmland and (b) linear waterways from models with year-specific first-year survival and constant adult survival (squares), year-specific adult survival and constant first-year survival (triangles) and year-specific first-year and adult survival (thin dashed line). Solid lines show the observed changes in abundance. Year-specific survival estimates are taken from model 1 in Table 1. All predictive models assume productivity to be constant over time.


Demographic conclusions

The evidence for a decline and subsequent increase in the survival rates of first-year reed buntings seems strong. Our analyses have demonstrated that the trend was not an artefact of any confounding trend in recapture probabilities, nor of any associated trend in the severity of winter conditions. The statistical evidence for a trend in adult survival was weaker but this may have been a consequence of less precise annual estimates (lower power) and possibly an underlying temporal trend of smaller magnitude. Although a recent analysis of UK ringing recoveries failed to detect any gross temporal changes in reed bunting survival rates (Siriwardena et al. 1998), this probably reflects lack of statistical power from a small data set. While it might be argued that changes in over-winter survival rates measured at four sites may not be representative of changes in the wider countryside, our ringing sites were widely spread across south and central England and reed buntings are known to disperse widely across farmland during winter. Although two-thirds of summer–winter UK ringing recoveries are within 5 km of the breeding site, the median distance of those moving further is 42 km (range 6–330 km) (Prys-Jones 1984). Studies in the vicinity of Brandon Marsh indicate that within-winter movements of 10–20 km are quite common (Fennell & Stone 1976). More than 99% of British breeding reed buntings are thought to remain within Britain during winter (Prys-Jones 1984).

We have also demonstrated that the timing and magnitude of the observed changes in both first-year and adult survival rates are sufficient, on their own, to account for much of the observed increase and subsequent decline in the British reed bunting population between 1969 and 1988 (Fig. 6). The temporal changes in reed bunting survival rates predicted a further substantial decline in numbers after 1987 that did not occur either on farmland or along linear waterways (Fig. 6). This departure of the predicted from the observed population indices, coupled with the evidence for lower nesting success since the mid-1980s (Figs 3 and 4), suggests that some other aspect of demography must have changed around that time. For example, there may have been an increase in the average number of nesting attempts or in post-fledging survival rates, neither of which was measured in this study.

Relatively high nesting success in both farmland and wetland habitats during the period of most rapid population decline suggests that nesting success per breeding attempt was not an important factor driving the population decline of the late 1970s. The decline in nesting success in northern Britain might have contributed to the larger, more prolonged population decline in that region (Fig. 2) and the decline in abundance in the south-east might have been more pronounced if nesting success had not been relatively high during the mid-1970s (Fig. 4). We acknowledge that our conclusions about the possible influence of changes in productivity on population levels are subject to the caveat that nest record cards cannot be used to measure changes in the number of breeding attempts per pair per season, and we cannot rule out the possibility that changes in this parameter may have contributed to the observed population changes.

Environmental factors that might have caused the observed changes in numbers and demography

It seems unlikely that climatic factors could have caused the decline in reed bunting numbers in Britain. Although severe winter weather did cause a dramatic population decline in 1963, and may have contributed to smaller declines in 1977, 1979 and 1982 (Fig. 1), other cold winters had little apparent effect on numbers (e.g. 1969–70, 1985–86 and 1990–91) and substantial declines were recorded in years following mild winters (e.g. 1978 on farmland and 1976, 1980 and 1981 along linear waterways, Fig. 1). Furthermore, the trend in first-year survival identified in this study does not seem to have been caused by any associated changes in the severity of winter weather (Table 1).

Loss and degradation of preferred wetland breeding habitat is likely to have contributed to the decline of British reed buntings. Extensive land drainage (Thomas, Allen & Grose 1981; NCC 1984) and the dredging and straightening of rivers (Smith 1975; Campbell 1988) are likely to have resulted in reduced breeding densities of reed buntings in many parts of Britain. However, the loss and degradation of wetland habitats has probably occurred continuously during the 20th century and it is not clear how the loss of such habitat could have caused the relatively sudden population decline between 1976 and 1983.

The most likely cause of the decline in the British reed bunting population is the loss of suitable food and habitat on farmland as a consequence of changes in agricultural practices. Half of all British reed buntings breed on farmland (Gregory & Baillie 1998) and farmland is probably their most important habitat during winter (Prys-Jones 1977; O’Connor & Shrubb 1986). The loss of hedges, ditches, ponds and uncultivated field margins coupled with the increased usage of herbicides probably removed much suitable reed bunting breeding habitat on farmland during the late 1970s. Reed buntings feed on a mixture of invertebrates and small weed seeds during summer but concentrate on small grass and weed seeds during winter (Prys-Jones 1977; Wilson, Arroyo & Clark 1996). The availability of both seeds and invertebrates on farmland is likely to have declined markedly since the mid-1970s with the introduction of a range of highly efficient insecticides and herbicides (O’Connor & Shrubb 1986; Donald 1998). Throughout the 1970s and early 1980s there was a large-scale switch from spring-sown to autumn-sown cereals in Britain that caused a large reduction in the availability of winter stubbles (Barr et al. 1993; Donald 1997; Shrubb 1997). Weed-rich stubbles have been shown to be strongly preferred as winter feeding habitat by reed buntings (Wilson, Taylor & Muirhead 1997), and by other buntings (Donald & Evans 1994; Evans & Smith 1994). The loss of winter stubbles was particularly marked between 1977 and 1984, during which time reed buntings declined by 53% both on farmland and along linear waterways (Fig. 7). Although the decline in the abundance of reed buntings is strongly correlated with the loss of spring-sown cereals on UK farmland (rs = 0·72, n = 26, P < 0·001 for the farmland index; rs = 0·86, n = 20, P < 0·001 for linear waterways), both variables show strong temporal trends and the correlations may therefore be spurious. Furthermore, despite the continued loss of spring-sown cereals since 1983, reed bunting numbers have remained relatively stable (Fig. 7).

Figure 7.

Changes in the abundance of reed buntings breeding on farmland (–▪–) in relation to changes in the area of UK spring-sown cereals ((–▾–) total area of wheat, barley and oats taken from Home-Grown Cereals Authority 1994).

The observed reduction in the survival rates of fully grown birds would be the predicted demographic consequence of a shortage of food during winter, and the apparently larger reduction in first-year survival may reflect a competitive advantage of adult over first-year birds. An increasing trend in the usage of gardens by reed buntings during winter (Thompson 1988) provides further circumstantial evidence of increasing food shortage on farmland, particularly during February and March when garden usage peaks (Cannon 1998).

The higher survival rates of young reed buntings during the 1990s (Fig. 4) might be a consequence of recent increases in winter food availability, perhaps through the provision of large areas of rotational set-aside in the British countryside (Wilson et al. 1995, 1997; Evans 1997a). However, these recently elevated survival rates relate to a much reduced population size, and the absence of even a partial recovery in breeding numbers (Crick et al. 1998) suggests that the carrying capacity of the winter environment is still much lower than it was during the early 1970s. A reduction in the abundance of invertebrates during summer might account for the lower nesting success of reed buntings in both farmland and waterside habitats since the mid-1980s, but this trend requires further investigation.

It is unclear why reed buntings might have experienced larger population declines and reduced nesting success in northern Britain. Our information on survival relates to southern Britain and it is possible that trends in survival have been different in northern Britain. The relatively large decline in abundance on northern farmland coincided with a contraction of breeding range in northern and western areas of Britain (Gibbons, Reid & Chapman 1993). Lower densities of reed buntings in the north and west, coupled with a more patchy breeding distribution (Marchant et al. 1990; Gibbons, Reid & Chapman 1993), might render such populations more vulnerable to local declines and extinctions (Harrison 1991), particularly following weather-related nesting failures or winter mortality. Losses of breeding range for the corn bunting have been proportionately greater in the more fragmented and lower density parts of the range, which, as for the reed bunting, tend to be concentrated in the northern and western parts of Britain (Donald & Evans 1995).

Given that reed buntings appear to prefer wetland to dry agricultural habitat and probably expanded into drier habitats when population levels were high (Bell 1969; O’Connor & Shrubb 1986), why was the decline on farmland not greater than that along linear waterways? It may be that the population decline that began in the late 1970s was mediated partly through a decline in the carrying capacity of winter farmland (as reflected in reduced over-winter survival) and partly through a loss of suitable breeding habitat both on farmland and along waterways. If the decline had been mediated solely through a reduction in the carrying capacity of farmland during winter, then we might predict that the rate and magnitude of decline should have been greater on farmland than along waterways.


The results of this and other studies (Evans & Smith 1994; Donald & Evans 1995; Evans 1997b) are consistent with the hypothesis that major declines in the availability of grass and weed seeds on British farmland during winter are likely to have been a major cause of the large population declines experienced by several species of granivorous passerine since the mid-1970s. When weed-rich winter stubbles were provided within the restricted range of the British cirl bunting population, numbers increased from approximately 120 pairs to 350 pairs in just 4 years (Evans 1997a). Although experimental research is now required to confirm the hypothesis that the availability of winter seed limits the survival rates and population sizes of granivorous farmland birds, the existing evidence is already strong enough to advocate changes in agricultural practices that increase the abundance of seeds on farmland as an appropriate conservation measure for these birds.

Organically managed farms are likely to provide more seeds (Moreby et al. 1994) and probably support more seed-eating birds in winter (Chamberlain, Wilson & Fuller 1999), but are unlikely to constitute more than a tiny proportion of the total cultivated land area in Britain in the foreseeable future. Rotational set-aside created relatively large areas of winter stubbles during the early 1990s and this is known to have been utilized by a range of declining granivorous birds in winter (Evans 1997b; Buckingham et al. 1999). However, with the possible exception of linnet, there was no general increase in the UK populations of most seed-eating birds during the peak set-aside era (Crick et al. 1998), suggesting either that insufficient winter food was provided at the national scale or that bird populations may now be limited by other factors. The probable disappearance of set-aside from European Union countries during the next few years will only increase the urgent need for agricultural policies that provide seed-rich habitats. Extending existing agri-environment schemes such as the Countryside Stewardship and the Pilot Arable Stewardship Schemes, which make specific provision for seed-rich habitats, is one potentially useful policy mechanism in this respect.

The recent decline in the nesting success of reed buntings suggests that conservation measures aimed at improving breeding productivity should also be considered. Measures that might increase productivity on farmland include the creation of reduced-input ‘conservation headlands’ to increase invertebrate prey (Stoate, Moreby & Szczur 1998), the maintenance and creation of ponds and wet ditches with fringe vegetation for nest sites and invertebrate prey and, on oilseed rape, the use of desiccating herbicide sprays, rather than swathing prior to harvesting, should reduce the numbers of nests destroyed by cutting machinery (Burton 1998). Measures to increase productivity in wetland require further research but might include the provision of rank vegetation and scrub as nest sites on the edges of water bodies and ditches, and the creation of shallow margins with emergent vegetation to promote invertebrate abundance.


We are indebted to the many hundreds of BTO volunteers who collected the data on which this analysis is based. We are particularly grateful to Dave Stone and the Brandon Ringing Group, Cyril Matthews and the Chew Valley Ringing Station, the late Bob Spencer (who initiated and ran the Marsworth Reservoir study for many years), and Dr Chris Thorne and the Wicken Fen Ringing Group for providing the mark–recapture data. W. J. Peach’s and R. D. Gregory’s posts were partly funded through a contract with the Joint Nature Conservation Committee on behalf of English Nature, Scottish Natural Heritage, the Countryside Council for Wales and the Environment and Heritage Service in Northern Ireland. G. M. Siriwardena was funded by the UK Ministry for Agriculture, Fisheries and Food under contract BD0906 to the Ecology and Behaviour Group University of Oxford. Dr Chris Wernham checked the recent data on seasonal movements and Dr Åke Berg and an anonymous referee provided helpful comments on an earlier draft of the manuscript.


Appendix 1

Farm type definitions

Arable: all fields used for tillage. Grazing: no crops grown and grass fields used for grazing or silage. Mixed: farms with both tillage and grazing.

Region definitions

Three regions were defined using aggregations of neighbouring Nomenclature of Territorial Units for Statistics (NUTS) regions (for a detailed map see Gregory & Marchant 1996). South-east Britain: south-east England, East Anglia and the East Midlands (NUTS regions 73, 74 and 75). South-western Britain: south-west England, Wales and the West Midlands (NUTS regions 76, 77 and 79). Northern Britain: rest of northern England and Scotland (NUTS regions 71, 72, 78, 7 A and 7B).

Appendix 2

Numbers of nest record cards contributing to various analyses

Temporal analysesRegional analysesHabitat analyses
Time period*Long declineShort declineNorthSouth-EastSouth-WestFarmlandWaterside
  • *

    The mean annual sample size of nest record cards was 106·5 (SE 4·1).

  • Time periods for the short decline analyses were: 1962–75, 1976–78 and 1979–95.


Test statistics for analyses of nest record card data

All test statistics and P-values below refer to likelihood-ratio tests between models allowing the appropriate variable to vary between the categories or interaction specified, and models in which the variable was removed. Significant results from the tests without interactions are accompanied by the estimates derived for each category.

Tests for temporal variation

Long decline period ( inline image, P): CS, 0·75, 0·69; BS, 0·94, 0·62; CER, 14·16, < 0·001 (1962–74, 0·931; 1975–83, 0·915; 1984–95, 0·897); EFR, 24·25, < 0·001 (1962–74, 0·010; 1975–83, 0·009; 1984–95, 0·021); NFR, 0·29, 0·86.

Short decline period ( inline image, P): CS, 7·11, 0·03 (1962–75, 4·47; 1976–78, 4·37; 1979–95, 4·53); BS, 2·96, 0·23; CER, 14·39, < 0·001 (1962–75, 0·933; 1976–78, 0·917; 1979–95, 0·899); EFR, 9·63, 0·008 (1962–75, 0·0106; 1976–78, 0·0097; 1979–95, 0·0208); NFR, 1·05, 0·59.

Tests for regional variation

Region effects ( inline image, P): CS, 0·73, 0·69; BS, 5·82, 0·055 (North, 4·12; South-East, 4·04; South-West, 4·18); CER, 3·79, 0·15; EFR, 1·15, 0·56; NFR, 6·18, 0·046 (North, 0·0194; South-East, 0·0260; South-West, 0·0315).

Region×time interaction ( inline image, P): CS, 6·72, 0·15; BS, 12·19, 0·016; CER, 1·95, 0·75; EFR, 2·89, 0·58; NFR, 7·95, 0·093.

Tests for habitat-specific variation

Habitat effects ( inline image, P): CS, 2·07, 0·15; BS, 0·35, 0·55; CER, 0·67, 0·41; EFR, 0·33, 0·57; NFR, 0·72, 0·40.

Habitat×time interaction ( inline image, P): CS, 4·86, 0·088; BS, 6·7, 0·035; CER, 7·13, 0·028; EFR, 3·43, 0·180; NFR, 0·10, 0·95.

Sample sizes for mark–recapture survival analyses

First caught as adultsFirst caught as juveniles
SiteNo. birds marked (1969–96)No. retrapped in subsequent yearsNo. birds marked (1969–96)No. retrapped in subsequent years
  • *

    The Wicken Fen data for birds first marked as adults were not available in computerized form.

Brandon Marsh9241311312129
Chew Valley Lakes588811380121
Marsworth Reservoir4468834542
Wicken Fen*76358