existing population estimates
These data were derived using four different methods. For the majority of species (51 of 92), territory mapping data from 144 CBC plots surveyed in 1989 were the basis (Marchant et al. 1990). Estimates of density were generated for three habitat types: woodland, farmland and ‘other habitats’. Densities in the first two habitat types were reported by the CBC; estimates for other habitats were the average of woodland and farmland densities weighted by the proportion of these habitats occupied. Estimates were corrected for geographical bias, using data on the number of 2 × 2 km squares (tetrads) occupied per 10 × 10 km square obtained from the second breeding bird atlas (Gibbons et al. 1993).
For 18 species, population estimates were derived from the number of 10 × 10 km squares occupied during surveys for the second breeding bird atlas. Data are based on two 1-h counts during early and late parts of the breeding season, to at least eight tetrads within each 10 × 10 km plot and supplemented with opportunistic records. The number of occupied 10 × 10 km squares was multiplied by an informed estimate of pairs per occupied square.
For six species, counts recorded during the second breeding bird atlas were used to estimate population density, which was multiplied by the area of the UK to obtain the population estimate. As tetrads were not selected randomly, population densities from randomly located point counts in approximately 15% of 10 × 10 km squares were used to correct for variation in densities in tetrads that were and were not visited.
For the remaining 17 species, estimates were obtained from species-specific information. Some were derived from targeted national surveys (e.g. mute swan Cygnus olor), which we believe to be of good quality. Others were essentially guesswork (e.g. feral pigeon Columba livia).
bbs-based population estimates
For all comparisons, new estimates were derived from BBS data collected in 2000 to be comparable with estimates in Birdlife International (2004). Because of its applied value, we also present population estimates derived from BBS data using the same methods for 2006, the most recent year for which data were available at the time of writing this study. The BBS is carried out annually on a stratified random sample of over 2500 1-km squares. BBS squares are stratified regionally by human population density in order to make the most of volunteer resources. These regions are referred to as sampling regions hereafter. Fieldwork involves two visits to each square, the first between early April and mid-May and the second between mid-May and late June. Birds are recorded in each 200-m section along two 1-km transects. Their perpendicular distances from the transect line are recorded in three distance bands (0–25 m, 25–100 m, 100 m or more). The number of distance bands is a compromise between collecting useful data and the ease with which volunteers can carry out the survey. However, even with a few bands, resulting population estimates will be unbiased provided that birds are allocated to the correct band (Buckland et al. 2001). The number of BBS squares recording each species, number of observations of each species, and the percentage of observations in the 0–25 m band in 2000 are shown in Table S1 (Supplementary material). Flying birds actively using resources in a square, such as displaying skylarks Alauda arvensis, are assigned to the appropriate distance band; records of flying birds are excluded from analyses.
Observers exclude juvenile birds from counts where possible because their inclusion would inflate estimates of the breeding population. Such errors are more likely for species that breed early, mainly residents, and those counted during the late visit. Conversely, most long-distance migrants are still arriving in the UK during the first visit period; thus, counts of these species during this visit may underestimate densities. To control for these biases, we only use data from early visits for resident species and only data from late visits for migrants. The exception is corn bunting Emberiza calandra, a late-breeding resident, for which we used data from the late BBS visit only (May to late June).
For the colonial nesting rook Corvus frugilegus and sand martin Riparia riparia, counts of active nests are also recorded. For these species, we use the maximum nest count multiplied by 2 to be equivalent to the number of adults, where this value exceeds the count of individual birds.
The main habitat within each 200-m transect section was recorded by observers according to Crick (1992).
Detectability must be taken into account when converting counts into densities, and subsequently, population estimates. This was achieved using distance sampling software (distance 5·0; Thomas et al. 2006). Birds recorded in the final distance band (100 m or more) were excluded from analyses, because counts within an unbounded category are difficult to interpret.
Whether birds are in flocks is not recorded in the BBS, and the majority of British species tend not to form flocks during the breeding season. Therefore, we assume that counts are of individual birds for analyses. Even for species that sometimes flock during the breeding season, such as the barn swallow Hirundo rustica, this approach is likely to yield a reliable population density estimate if these four conditions are met: (i) all flocks on the transect line are detected; (ii) mean flock size is not very large; (iii) flock size is not highly variable; and (iv) detectability of flocks is not very low (Buckland et al. 2001). For species in this study, with the exception perhaps of common linnet Carduelis cannabina, these conditions appear likely to be met.
Habitats differ in their structure which is likely to influence detectability. Detectability may also vary between regions for a variety of reasons. The structure of a single habitat may differ with geographical location (e.g. denser woodlands in southern UK), for example. Population densities may differ between regions, and in areas with high densities, individual birds may need to invest more time in territory defence, singing more frequently and increasing their detectability. We took heterogeneity in detectability arising from variation between habitats and regions into account by including these as factor covariates in a global multiple-covariate detection function (Marques & Buckland 2003). This approach uses the whole data set to provide information about the shape of the detection function, so it was not necessary to fit a separate detection function to each habitat–region combination.
Following Buckland et al. (2004), we fitted half-normal and hazard-rate distributions to the data. We identified nine main habitat types (broad-leaved, coniferous and mixed woodland, semi-natural grassland, heath and bog, arable, pastoral and mixed farmland, human habitats and water bodies), and 12 regions (nine English Government Office Regions, Wales, Scotland and Northern Ireland) and adopted the following stepwise approach. For each species, we estimated f(0), i.e. the value of the probability density function of perpendicular distances at zero distance, without and including one and then both habitat or region as covariates to both half-normal and hazard rate models. Akaike's Information Criterion (AIC) was used to identify the best-fitting model for each species. This model was then applied to the encounters from surveyed squares to produce an estimate of individuals within each square. Whilst habitat was retained as a covariate for all species, for a quarter of species (23 of 92), region was also retained. This suggests that even when habitat effects are taken into account, detection functions may vary spatially.
To obtain estimates of population size and confidence intervals, we used a bootstrap resampling procedure of 399 iterations (Crowley 1992). For each iteration, 1 km squares were randomly resampled with replacement within the sampling region, the sample drawn being equal to the total number of 1 km squares in that region. The values for each sampling region were summed to derive an estimate of each species in that region and region totals summed to give an estimate of total population. The 10th and 390th ordered bootstrap values across iterations and total population estimate were taken to give the lower and upper 95% confidence limits of the national estimates, respectively.
Most estimates presented in BirdLife International (2004) were of breeding pairs. For comparative purposes, we multiplied these by 2 to get estimates of individual birds. Statistical comparisons in this study are based on log10-transformed population size estimates.
We expected BBS-based estimates to be positively correlated with previous estimates, and the relationship to be monotonic. However, we had no general expectation as to the strength of the correlation or exact form of the relationship.
We categorized each species according to these five broad habitat categories: (i) woodland; (ii) human habitats (urban, suburban and rural); (iii) farmland (arable and pastoral); (iv) semi-natural grassland, heath and bog; and (v) wetlands (standing and flowing water). We used a smaller number of habitat types than that used to model habitat differences in detectability because (i) habitat types were maximized in detectability modelling to increase precision in taking detectability differences into account, and (ii) the four woodland categories used to model detectability were merged, as our primary aim was to distinguish between species whose characteristic habitat types corresponded to those covered well by the CBC (i.e. woodland and farmland). We took the proportion of BBS 200-m sections in each class as the availability of that class (p) and the proportion of sections on which a species occurred that were in that class as the species usage of that class (r). We calculated Jacobs (1974) preference index of the species for each habitat:
J = (r – p)/[(r + p) – 2rp]
The habitat with the largest index for that species was taken to be its characteristic habitat, see Table S1.
Previous population estimates for 51 common farmland and woodland species were based on CBC territory mapping and are likely to be fairly accurate. The BBS-based population estimates for such species are also likely to be unaffected by marked bias, and we thus expect the two sets of estimates to be similar. Conversley, previous estimates for species whose main habitat type is neither farmland nor woodland are less likely to be accurate. For these species, estimates are based on population densities in farmland and woodland, which are not the focal species’ preferred habitat, and this may lead to their national population size being underestimated. For such species, we expect BBS-based estimates to be higher than previous estimates.
Estimates derived from CBC data are based on the number of territories, whilst BBS-based estimates concern the number of adult birds. For all species, there are likely to be non-territory holders that will not be recorded by the CBC (Newton 1998). This is likely to be greatest for species which delay breeding until their second year or older, and thus have a large proportion of non-breeding individuals in their population. Therefore, we expect that BBS estimates will be larger than previous estimates for species which delay breeding. Using the ornithological literature (Cramp, Simmons & Perrins 1977–1994), we classified species as those in which the majority of adults start breeding from 2 or more years of age, 3 or more years, and where the majority of individuals start breeding in their first year (see Table S1).
Counts made during timed tetrad visits may have missed some birds, whilst estimates derived from BBS data explicitly use distance sampling to take undetected individuals into account, giving higher estimates.
Whilst previous estimates based on the number of occupied squares are of lower quality than those derived from CBC or count data (Stone et al. 1997), we did not expect the former, or previous estimates based on species-specific methods, to differ from BBS-based estimates in any systematic way.
The CBC was geographically biased towards southern England. Within a rectangle of easting 3000 and northing 5000 of the National Grid, which we refer to as ‘Southern UK’, the CBC farmland plots were representative of the general landscape (Fuller, Marchant & Morgan 1985). Because we have produced estimates of population size for sampling regions, we use these to estimate the proportion of the UK populations of each species lying within and outside Southern UK. If corrections for geographical bias applied to the CBC data were under- or overcompensated for, UK population estimates for species predominantly outside Southern UK would be systematically biased when based on BBS data. To test this, we examined two groups: species for which > 60% and 70% of the population lie outside Southern UK (see Table S1). The latter group is more likely to be biased, but the former group provides a larger sample.
Effects of sex on detectability
The BBS does not ask observers to record sexes separately even if they are identifiable by plumage or through differences in behaviour (e.g. singing/displaying males). For species where there is a significant difference in detectability between the sexes, analysing the data for the two sexes combined may underestimate true population size. Simulations carried out by the authors to examine the influence of such differences in detectability and work by Buckland et al. (2001) have shown that distance sampling is robust to fairly large sex-based differences in detectability. However, this breaks down at levels where detection is mainly of one sex. For this reason, using the ornithological literature and an independent assessment by experienced field ornithologists, we focus on species for which sex bias in detectability is likely to be extreme (see Table S1). We expected that BBS-based population estimates would be smaller for species with a marked difference in detectability between sexes.
We used paired t-tests to examine whether BBS and previous estimates were significantly different from one another for species associated with each of the five habitat classes. We used Generalized Linear Models to test for the effects of previous survey methods, delayed breeding, geographical bias and influence of sex on detectability on the difference between BBS and previous estimates, defined as log10(BBS/previous). Multiple regression analysis was used to quantify the extent to which habitat, previous survey method, delayed breeding, geographical bias and influence of sex on detectability explain differences between BBS and previous estimates. For these analyses, an information theoretic approach to model selection was used (Burnham & Anderson 2002). All analyses were carried out in sas (SAS Institute 2001).