Abundance–occupancy dynamics in a human dominated environment: linking interspecific and intraspecific trends in British farmland and woodland birds



    1. Department of Animal & Plant Sciences, University of Sheffield, Sheffield S10 2TN, UK; British Trust for Ornithology, The Nunnery, Thetford, Norfolk IP24 2PU, UK
    Search for more papers by this author

    1. Department of Animal & Plant Sciences, University of Sheffield, Sheffield S10 2TN, UK; British Trust for Ornithology, The Nunnery, Thetford, Norfolk IP24 2PU, UK
    Search for more papers by this author

    1. Department of Animal & Plant Sciences, University of Sheffield, Sheffield S10 2TN, UK; British Trust for Ornithology, The Nunnery, Thetford, Norfolk IP24 2PU, UK
    Search for more papers by this author

Thomas J. Webb, Department of Animal & Plant Sciences, University of Sheffield, Sheffield S10 2TN, UK. Tel.: +44 1142220057; fax: +44 1142220002; e-mail: t.j.webb@sheffield.ac.uk


  • 1Range size, population size and body size, the key macroecological variables, vary temporally both within and across species in response to anthropogenic and natural environmental change. However, resulting temporal trends in the relationships between these variables (i.e. macroecological patterns) have received little attention.
  • 2Positive relationships between the local abundance and regional occupancy of species (abundance–occupancy relationships) are among the most pervasive of all macroecological patterns. In the absence of formal predictions of how abundance–occupancy relationships may vary temporally, we outline several scenarios of how changes in abundance within species might affect interspecific patterns.
  • 3We use data on the distribution and abundance of 73 farmland and 55 woodland bird species in Britain over a 32-year period encompassing substantial habitat modification to assess the likelihood of these scenarios.
  • 4In both farmland and woodland habitats, the interspecific abundance–occupancy relationship changed markedly over the period 1968–99, with a significant decline in the strength of the relationship.
  • 5Consideration of intraspecific dynamics shows that this has been due to a decoupling of abundance and occupancy particularly in rare and declining species. Insights into the intraspecific processes responsible for the interspecific trend are obtained by analysis of temporal trends in the distribution of individuals between sites, which show patterns consistent with habitat quality declines.
  • 6This study shows that a profitable approach to ascertaining the nature of human impacts is to link intra- and interspecific processes. In the case of British farmland and woodland birds, changes to the environment lead to species-specific responses in large-scale distributions. These species-specific changes are the driver of the observed changes in the form and strength of the interspecific relationship.


The extent to which humans have modified the Earth's ecosystems is well documented (e.g. Daily 1995; Vitousek et al. 1997; Chapin et al. 2000; Gaston 2004). Consequences have included the extinction or large-scale decline of many species, as well as (frequently intentionally) increases in others (Vitousek et al. 1997; Chapin et al. 2000). As a result the large-scale abundances and distributions of many species across large areas of the Earth have changed significantly. To the extent that such increases and declines are not partitioned randomly across taxa (Bennett & Owens 1997; Russell et al. 1998; Lockwood, Brooks & McKinney 2000), these effects also modify the distributions of other biological traits such as body size and hence are likely to have profound implications for the ways that whole ecosystems function (Chapin et al. 2000).

Macroecologists have devoted considerable attention to studying the abundance and distribution of species at large scales, as well as the structure of ecological or taxonomic assemblages, for example in terms of their body size distributions (Brown 1995; Gaston & Blackburn 2000). However, despite the clear expectation that the primary macroecological variables (abundance, range size, body size) will all be affected by human activity, this issue has received surprisingly little attention in the macroecological literature (Gaston & Blackburn 2003a; Gaston 2004). Nevertheless, the few macroecological studies that have, directly or indirectly, incorporated human activity have revealed large effects. For instance, range size–body size relationships in Australian mammals have been substantially altered by extinctions and artificial range reductions over the past two centuries (Murray & Dickman 2000), and direct human exploitation has led to a dramatic under-representation of large individuals in the North Sea fish fauna (Jennings & Blanchard 2004).

Human activity therefore has detectable effects on the frequency distributions of abundance, range size and body mass of contemporary taxa, but its effects on more complex macroecological patterns (for instance, the relationships between such variables) are harder to predict. One of the most important large-scale patterns is the generally positive relationship between local population density and site occupancy, the abundance–occupancy relationship. This pattern has been documented across a broad range of taxa in many habitats (see Gaston et al. 2000 for a review). Abundance–occupancy relationships are important because they provide a link between local and regional scale population ecology (Freckleton et al. 2005), because they potentially result from changes in the quality of habitat and its distribution (Holt et al. 1997; Freckleton, Noble & Webb 2006), and because correlated declines in abundance and occupancy suggest a ‘double jeopardy’ for species declining as a consequence of human activity (Lawton 1995).

In general, despite the importance of interspecific abundance–occupancy relationships, and the large amount of interest they have generated (see Gaston et al. 2000 and references therein), relatively little attention has been paid to their temporal dynamics. One exception is a study of 24 marine fish species from eastern Canada (Fisher & Frank 2004). Here, the slope of the interspecific abundance–occupancy relationship showed a marked temporal trend over a 32-year period, as the abundance and distribution of individual species changed in response to fishing activity. An earlier study of British breeding birds (Blackburn et al. 1998) suggested that the abundance–occupancy relationship remained relatively constant over time, and that individual species rarely changed their rank order in either abundance or occupancy. However, these authors did not investigate temporal trends in the form or strength of the abundance–occupancy relationship.

In the context of human-induced changes in abundance, one question that remains unanswered is whether abundance–occupancy relationships are the same for species that are increasing in abundance as they are for those that are decreasing. As outlined below, there are a number of reasons to believe that the relationship between abundance and occupancy may be different for these two groups of species, but there have been to date no studies addressing this issue. Understanding such differences is important, as they would imply that the rate of change in population size resulting from habitat improvement is different from that resulting from habitat loss. To understand these effects requires that the underlying mechanics of population dynamics and structure are known.

Here, we examine the interspecific abundance–occupancy relationship in British birds breeding on farmland and woodland during a period of profound anthropogenic change in order to determine how changes to the environment might affect macroecological relationships. We begin by outlining what patterns of temporal dynamics might be expected as habitats and species densities change. We test explicitly for temporal trends in both the strength and the form of these relationships. We then relate these trends to intraspecific dynamics. Specifically, we consider differences between rare and common species, and between increasing and declining species. Finally, we discuss potential mechanisms behind the observed patterns in terms of changes in the distribution of individuals across space.

expected temporal dynamics in abundance–occupancy relationships

Numerous studies have explored the form of intraspecific abundance–occupancy relationships, that is the covariation of abundance and occupancy over time within species (see Gaston et al. 2000 for review). In addition, Blackburn et al. (1998) examined temporal trends in the interspecific relationship for British birds, Fisher & Frank (2004) suggested links between intra and interspecific dynamics, and Gonzalez et al. (1998) recorded changes in the interspecific relationship in response to experimental habitat fragmentation in microarthropod communities. However, links between intra- and interspecific processes have seldom been addressed. In Fig. 1, we illustrate four possible forms that such interactions may take.

Figure 1.

Potential trajectories for the interspecific abundance–occupancy relationship (solid line in each panel) as intraspecific densities and occupancies change (arrows). (a) Density and occupancy are tightly coupled in all species; species move up or down the existing relationship, which remains essentially unchanged in terms of its parameters. (b) Changes in density are inversely related to changes in occupancy; if all species increase in density and therefore decrease in occupancy (or vice versa) the resulting interspecific relationship will have a new intercept but other parameters (slope, shape, correlation) will remain constant (dashed lines). A mixture of declining and increasing species will modify these parameters. (c) Intraspecific density–occupancy relationships follow no consistent pattern; density and occupancy are essentially unrelated. Changes in one or other variable across multiple species will tend to lead to a decrease in the interspecific density–occupancy correlation; effects on other parameters (slope, intercept, shape) will be unpredictable. (d) A combination of A and B or C. In this example, commoner species tend to conform to the interspecific pattern, rarer species do not. Such a situation will generally lead to a reduction in the interspecific density–occupancy correlation, but its effects on the other parameters may not be predictable.

The simplest expectation is that the intra- and interspecific relationships are the same, and that changing abundance simply results in species moving along a single curve (Fig. 1a; see Gonzalez et al. 1998). According to this scenario, there is no systematic deviation of intraspecific patterns from the interspecific relationship with changing density, and the correlation between abundance and occupancy is the same for all densities, irrespective of whether species are increasing or declining.

Second, intraspecific abundance–occupancy relationships may tend to be negative if the lowest density populations of a species are most likely to go extinct such that a decline in occupancy is associated with an increase in mean density across remaining occupied sites, or if species expand their distributions through the founding of new, low density populations. This would result in intraspecific relationships quite different from the interspecific pattern (Fig. 1b). An important prediction from this model is that the position (i.e. the intercept) of the relationship between occupancy and abundance may change markedly as a result of simultaneous increases or decreases in all populations (the dashed lines in Fig. 1b). Conversely (not shown), intraspecific relationships may be positive, but different from the interspecific one.

Alternatively responses may be species-specific, with the effects of environmental change on the interspecific abundance–occupancy relationship depending upon the relative frequencies of the above scenarios, as well as whether each scenario is equally likely for all species. In this case the overall relationship would be a composite (Fig. 1c). The shape of the intraspecific relationship should be unaffected by a species’ density, and the average correlation between abundance and occupancy across species should not change systematically over time.

Finally, intraspecific abundance–occupancy relationships might vary systematically with density (Fig. 1d). Two factors have been identified as being key drivers of abundance–occupancy relationships, namely colonization rate and the distribution of suitable habitat (Holt et al. 1997; Freckleton et al. 2005, 2006). These factors are likely to vary between relatively rare and common, and between increasing and declining species. Thus, abundant, high occupancy species would have high colonization rates and be able to respond rapidly to changes in local habitat quality, resulting in positive intraspecific relationships. On the other hand, species with low density and occupancy may have lower colonization ability (due to isolation of existing populations), occupy proportionately less suitable habitat than common species (e.g. Storch & Sizling 2002), respond less rapidly to change, and consequently show weaker intraspecific relationships (Fig. 1d). Moreover, modelling work has suggested that, under such circumstances, the variance about the abundance–occupancy should be greater at low densities than at high densities (in contrast with simple models of aggregation; Freckleton et al. 2006).

Some of the scenarios illustrated in Fig. 1 seem more likely than others. For instance, intraspecific abundance–occupancy relationships generally are positive more often than not (Gaston et al. 2000; Gaston 2003), so scenarios requiring primarily negative or no intraspecific abundance–occupancy relationships (Fig. 1b,c) are probably unlikely. Mixtures of simple models, perhaps with systematic differences between rare and common species (e.g. Fig. 1d), or between declining and increasing species, are perhaps more likely.



We used data on the distribution and abundance of British bird species from the Common Birds Census (CBC), collected by the British Trust for Ornithology (BTO) between 1962 and 2001. See Marchant et al. (1990) for full details of this data set, which has been used for previous analyses of abundance and occupancy (e.g. Blackburn, Gaston & Gregory 1997; Gaston, Blackburn & Gregory 1997a,b, 1998a; Blackburn et al. 1998; Gaston et al. 1998b; Freckleton et al. 2005, 2006). Briefly, the breeding birds on a large number of plots (c. 260 in any one year) throughout Britain were thoroughly censused on multiple occasions in each year between 1962 and 2001 using territory mapping. Supplementary data including the area and habitat type of each plot were also collected. Thus, reliable population estimates are available for each species in each year, and occupancy can be estimated as the proportion of all surveyed sites at which a species was recorded.

We restricted our analyses to surveys taken during the 32 years between 1968 and 1999, when standard methodologies were employed and the number of surveyed sites was consistently high. We used data from farmland and woodland plots (which each constituted about 35–50% of all CBC plots sampled in each year). These data cover a period during which British farmland bird populations have changed markedly, primarily as a response to changing farming methods (Chamberlain et al. 2000; Robinson & Sutherland 2002; Greenwood 2003). The CBC largely reflected these changes (Marchant et al. 1990), and has been used to highlight declines in farmland birds (e.g. DEFRA 2002). Woodland bird populations have also changed over the same period (Amar et al. 2006), at least partly in response to changing woodland habitats (e.g. less active management and increased deer activity; Amar et al. 2006). However, the link between habitat change and population change is less well established in woodland compared with farmland, and other factors (e.g. changes in the wintering habitats of long-distance migrants) are likely to have been important for woodland bird populations (Amar et al. 2006).

In keeping with previous analyses of CBC data we analyse the two habitats separately as they are not comparable in terms of plot size (mean annual plot area ranged from 74 to 86 ha on farmland, and from 21 to 27 ha on woodland) or typical densities attained (51 species in our data set bred on both habitats in each year; mean densities were on average (mean ± SD) 3·7 ± 1·43 times higher in woodland compared with farmland). This also allowed us to include species specialist in one or other habitat, as well as those that occur in both. The range of plot sizes within each habitat was relatively small, and Gaston, Blackburn & Gregory (1999) have shown that such small differences in census area have a negligible effect on species density estimates.

We estimated population density using counts only from sites deemed suitable for that purpose by the BTO. For each habitat, we considered only species that were recorded breeding on at least one plot in each year, and we excluded species (e.g. woodpigeon Columba palumbus L., rook Corvus frugilegus L.) for which population size was rarely censused in the early years of the CBC. This resulted in a sample size of 73 farmland and 55 woodland species.

interspecific dynamics

For each year, and separately for farmland and woodland habitats, we estimated proportional occupancy (the proportion of surveyed sites occupied) and mean population density (the mean density within occupied sites) for each species. We used these to estimate an interspecific abundance–occupancy relationship separately for each year, using several measures. First, we calculated the correlation of ln(density) and occupancy for each year. We also estimated the slope and intercept of a linear model of occupancy on ln(density). In both cases, using ln-transformed density improved the fit of the data to the assumptions of the tests. Finally, we extracted two parameters from a simple Generalized Additive Model (GAM; Hastie & Tibshirani 1990), which fit a nonparametric smooth function linking occupancy with density (untransformed). The parameters considered were the intercept of the model, and Estimated Degrees of Freedom (EDF). EDF is a measure of how many degrees of freedom's worth of smoothing are justified by the data; increasing its EDF shifts the form of a fitted smooth from a straight line (infinitely smooth, EDF = 1) through monotonic to increasingly multimodal. We used the mgcv package (Wood 2004) implemented in R (R Development Core Team 2004) to automatically select for each year the optimum EDF in the range 1–4. In all models we designated occupancy as the dependent variable, as it seems most likely that changes in occupancy will be driven by changes in abundance, rather than vice versa (Gaston & Blackburn 2003b). Trends in each of the estimated coefficients were quantified by their correlation with year.

intraspecific dynamics

In order to relate the trends in the interspecific abundance–occupancy relationship to intraspecific dynamics, we next estimated intraspecific abundance–occupancy relationships for each species, i.e. the relationship between density and occupancy within a species over time.

Before comparing intraspecific relationships across a group of species, it is important to recognize that these relationships will vary systematically with both density and occupancy. For instance, a species that occupies all sites in a region will not be able to increase its occupancy even if its density increases markedly, and so strong positive relationships would not be expected in such a species. It is possible, and in some circumstances entirely appropriate, to approximately linearize the interspecific relationship by transforming occupancy (e.g. using the logit transformation; Williamson & Gaston 1999). We did not do this because the decrease in the strength of the relationship at high occupancy is a real phenomenon that we wished to capture. We therefore fitted a nonlinear interspecific abundance–occupancy relationship and measure the intraspecific relationship of a species relative to this.

We obtained an overall interspecific relationship by fitting the model described by Freckleton et al. (2005) to the overall interspecific data (i.e. the mean density and occupancy for each species, averaged over all years). The model took the form:

image( eqn 1)

(see Freckleton et al. 2005 for details of how the model was fit). This model was chosen because it produced a good fit to the data (Fig. 2), which is the important factor in what follows. It is not intended to mechanistically describe the relationship. However, the parameters can be interpreted as follows. C is the slope relating occupancy to density at low density, g governs the shape of the relationship, changing the point at which the increase in occupancy with density becomes nonlinear and begins to approach an asymptote, and K allows for a nonzero intercept. In fitting, we weighted by the squared reciprocal of density to account for an expected decrease in variance with increasing density.

Figure 2.

The interspecific abundance–occupancy relationships for British farmland (a) and woodland (b) birds. Open symbols represent the mean density and occupancy of each single species, averaged over the period 1968–99. The correlation is highly significantly positive in both habitats (farmland: r = 0·74, d.f. = 71, P < 0·00001; woodland: r = 0·70, d.f. = 53, P < 0·00001). The solid line on each figure is the maximum likelihood fit of eqn 1 to the data, weighted by 1/density2. Solid symbols are running means of occupancy and density, to illustrate more clearly the good fit of the models.

To compare intra- and interspecific patterns, we used the fitted models to estimate the intraspecific relationship for each species relative to the interspecific relationship as follows. The fitted interspecific model (see above) was used to obtain predicted occupancy (ôinter) for each observed annual density estimate for a given species. These predicted values were then used as the null expectation for the intraspecific relationship by fitting a linear model of the form o ∼ d + offset(ôinter), where o and d are the observed annual occupancy and density estimates for the species. We term the coefficient returned from this model the ‘relative slope’ of the intraspecific relationship, because the offset term means that we are testing the slope of the relationship of o on d against a null expectation of the relevant portion of the interspecific curve. A relative slope of 0 indicates that the intraspecific relationship of the species is exactly parallel to the interspecific curve over the same range of densities. Positive and negative relative slopes, respectively, indicate relationships that are steeper or shallower than interspecific expectation. Note that a relative slope of 0 implies that intraspecific dynamics vary with density, as the interspecific relationship itself changes shape. This allows us to retain the advantages of linear models (e.g. the ease with which they can be summarized) while incorporating the nonlinear framework of the interspecific relationship.

As outlined above, the dynamics of species increasing in regional distribution may differ from those that are declining. Each species was classified as either declining or increasing on the basis of the temporal trend in its occupancy, which was assumed to capture the essential trajectory of its distribution over the period considered. A species was classed as declining if the correlation between occupancy and year was negative, and increasing if this correlation was positive. We compared these trends with an independent measure of distributional change: percentage change in the number of 10 km2 squares occupied across Britain (i.e. in all habitats) between the atlases of Sharrock (1976) and Gibbons, Reid & Chapman (1993). Across species, the two estimates of change were highly positively correlated in both habitats (farmland: rs = 0·69, d.f. = 70, P < 0·00001; woodland: rs = 0·57, d.f. = 53, P < 0·00001).

We used General Linear Models to examine the interspecific relationship between ln(mean density) and the relative slope of the intraspecific abundance–occupancy relationship. We included occupancy trend (declining or increasing) as a factor in this analysis, as well as the interaction between occupancy trend and ln(mean density).

changes in regional population structure

Changes in the interspecific abundance–occupancy relationship can potentially be related to changes in intraspecific regional population structure that have in turn resulted from changes in the distribution of habitat suitability, as well as from changes in average habitat suitability. We therefore examined for each species how the distribution of individuals across sites has changed over time, and related these trends to the species’ mean density, and to its occupancy trend. We measured the distribution of population densities for each species in each year as the skew of the distribution of densities, using the unbiased estimate of skew (g1) given in Sokal & Rohlf (1995). The skewness of a distribution measures its shape, not simply its spread, thus quantifies the extent to which population densities are biased towards being low (right skew, positive g1), high (left skew, negative g1), or symmetrically distributed around the mean (no skew, g1 = 0). We only calculated skew for years in which a species occurred at ≥ 10 sites, and trends in skew (the correlation of skew and year) were not estimated for species that had skew estimates for < 8 years (n = 9 species on farmland, 4 in woodland). We used General Linear Models to examine the interspecific relationship between ln(mean density) and trends in skew. We included occupancy trend (increasing or declining) as a factor in this analysis, as well as the interaction between occupancy trend and ln(mean density).

phylogenetic effects

We tested the data in order to determine whether a phylogenetic correction was required in the analysis, using the test of Pagel (1999) described in Freckleton, Harvey & Pagel (2002) and a phylogeny based on that of Sibley & Ahlquist (1990). We found no significant evidence of phylogenetic dependence in the data used in the analyses reported below, and hence we report across-species analyses only. The lack of phylogenetic dependence reflects large variations in abundance and occupancy even for very closely related species. Thus meadow pipits Anthus pratensis L., linnets Carduelis cannabina L., blue tits Parus caeruleus L. and blackbirds Turdus merula L. are all considerably more widespread on farmland than their respective congeners tree pipits A. trivialis L., redpolls C. flammea L., willow tits P. montanus Conrad and mistle thrushes T. viscivorus L. The lack of phylogenetic signal in variables describing the abundance and distribution of species appears to be rather a general phenomenon (Webb & Gaston 2003, 2005).


abundance–occupancy trends

The interspecific correlation between abundance and occupancy over the period 1968–99 declined markedly for both British farmland and woodland birds (Fig. 3; Table 1). In farmland birds this decline has been accompanied by a significant decline in the linear slope and intercept (Table 1), with GAMs showing that the relationship has also become apparently less curved over time (negative trend in EDF; Table 1). This is probably a consequence of the increasing scatter in the relationship, as extra degrees of freedom in the GAM will only be justified when the model fits the data well. Thus, the R2 from the fitted GAMs is positively related to EDF (r = 0·63), and is highly correlated with the abundance–occupancy correlation (r = 0·99). In woodland birds, there is no trend in GAM EDF (Table 1), possibly because the woodland relationship was rather more scattered than the farmland relationship, leading to generally lower EDFs (< 2 in 19 of 32 years). GAM R2 did decline over time, however (r = −0·75), and again was very tightly associated with the simple abundance–occupancy correlation (r = 0·96), unsurprisingly given that the GAMs did not depart markedly from linearity.

Figure 3.

Temporal trends over 32 years in the interspecific abundance–occupancy correlation in (a) farmland (r = −0·73) and (b) woodland (r = −0·59) birds.

Table 1.  Temporal trends in parameters describing the interspecific abundance–occupancy relationship. The correlation and the linear model used ln-transformed density; the GAM was on untransformed density. n = 32 years for each measure
Abundance–occupancy measureFarmlandWoodland
Trend over time (r)PTrend over time (r)P
Correlation−0·73< 0·00001−0·590·0003
Linear slope−0·53  0·0017 0·480·0056
Linear intercept−0·50  0·0034 0·550·0010
EDF from GAM−0·62  0·0001−0·170·3490
GAM intercept 0·27  0·1278 0·140·4441

The linear slope of the interspecific relation- ship actually increased over time in woodland birds (r = 0·48; Table 1), despite a decrease in correlation. Thus although the two variables are less closely associated in terms of the proportion of variance explained, the relationship has actually become stronger in terms of the unit change in occupancy with changing density in woodland birds. This is perhaps because variance in occupancy in woodland birds has increased over time to a much greater degree (r = 0·91, P < 0·00001) than has variance in density (r = 0·10, P = 0·5972). In terms of occupancy, common species have become more common, whereas rare species have become rarer. For instance, the number of species with occupancies < 0·125 doubled from eight in 1968 to 16 in 1999; at the same time so did the number of species with occupancies > 0·875 (from four in 1968 to nine in 1999). Mean occupancy over all years was significantly higher for increasing species (0·65 ± 0·064) than for declining species (0·28 ± 0·043; F1,53 = 33·1, P < 0·00001). In contrast, on farmland the trend in variance in occupancy was less marked (r = 0·52, P = 0·025). The number of species with very low or very high occupancy remained relatively constant (19 and nine species had occupancies < 0·125 and > 0·875, respectively, in 1968, changing in 1999 to 14 and seven species, respectively) and there was no difference in mean occupancy between declining (0·38 ± 0·050) and increasing species (0·43 ± 0·068; F1,71 = 0·45, P = 0·5049).

linking intraspecific and interspecific patterns

Potential reasons for the decline in the interspecific abundance–occupancy correlation can be uncovered by considering the intraspecific relationships. Figure 4(a,b) show the intraspecific relationships for all farmland and woodland species, with the overall interspecific relationships superimposed. Clearly, species tend to remain restricted to a portion of the interspecific relationship over time, i.e. common species remain common and rare species remain rare (see also Blackburn et al. 1998).

Figure 4.

The interspecific abundance–occupancy relationship, shown as the model fitted in Fig. 2 (thick line), for (a) farmland and (b) woodland birds. Here, instead of plotting a single time-averaged density and occupancy for each species, we plot for each species the linear relationship over time between its density and occupancy. Declining species are shown as solid lines, increasing species as dashed lines. Note that fitted values > 1 were changed to 1, the maximum possible occupancy. (c, d) The departure of the intraspecific abundance–occupancy slopes from the interspecific slope, across (c) farmland and (d) woodland species, as a function of mean density. Filled symbols and the solid line are declining species, open symbols and the dashed line are increasing species. Fitted lines are from a General Linear Model modelling the relative intraspecific slope as a function of ln(mean density) and occupancy trend, and the interaction between density and trend (see Table 2). The dotted horizontal line is at 0 (no trend).

It is also clear that there is considerable interspecific variation in the slope of the intraspecific relationships. However, significantly more than half were positive in both habitats (farmland: 50 of 73, binomial probability = 0·0005; woodland: 42 of 55, P < 0·00001). This suggests that the scenarios in Fig. 1(b,c), either of which could account for the observed reduction in the abundance–occupancy correlation, are both unlikely. Figure 1(a) seems equally unlikely, as it would predict a constant or increasing interspecific correlation. The fact that 20–30% of species have intraspecific correlations < 0, and that most species occur at low densities, i.e. in the region where the interspecific relationship is strongest, also runs counter to the predictions of Fig. 1(a). Thus, a mixture of intraspecific relationships has combined to produce the observed reduction in the interspecific correlation. The question then becomes whether intraspecific relationships vary systematically with mean density (Fig. 1d).

Substantially more intraspecific variance around the interspecific abundance–occupancy relationship appears to occur at low compared with high densities (Fig. 4a,b). This is confirmed by plotting the slope of each species’ intraspecific relationship (relative to the interspecific relationship over the same range of densities) against its mean density calculated over all years (Fig. 4c,d). We excluded one species from the farmland analysis: the sparrowhawk Accipiter nisus L. which, with a relative intraspecific slope of −42, is a conspicuous outlier in quantitative (but not qualitative) terms.

In both habitats, there is a strong positive correlation between ln(mean density) and relative abundance–occupancy slope (farmland: r = 0·56, d.f. = 70, P < 0·00001; woodland: r = 0·64, d.f. = 53, P = 0·00001): the most common species tend to conform well to interspecific expectation, whereas variation around this relationship is much more pronounced for rarer species, which tend to have less strongly positive relationships than predicted (Fig. 4c,d). Separating species by occupancy trend, there is a suggestion that the relationship between mean density and relative slope differs between the two groups of species on farmland, with a stronger relationship in declining species then in increasing species (Table 2). Forcing the same slope (4·20 ± 0·739) on both groups of species shows that increasing species conform more closely to the interspecific relationship than do declining species: the increasing species intercept exceeds the declining species intercept by 2·22 ± 1·067 (t = 2·09, P = 0·0408). In woodland, the slope of the relationship in declining species is significantly greater than that in increasing species (Table 2): the slopes of rare declining species fall furthest from interspecific expectation. These results provide broad support for the scenario outlined in Fig. 1(d).

Table 2.  Slopes of the relationship between ln(mean density) and relative intraspecific abundance–occupancy slope for declining and increasing farmland and woodland species
 Declining speciesIncreasing speciesDifference in slopes
Farmland345·49 ± 1·071393·08 ± 1·4641·640·1048
Woodland302·60 ± 0·437250·95 ± 0·5483·000·0041

changes in regional population structure

In both habitats, the temporal trend in the skew of population densities of a species was related to its mean density, and to its occupancy trend (Fig. 5a,b). Positive relationships between ln(mean density) and trend in skew were observed in declining species, whereas increasing species showed the opposite trend (Table 3). The implications of these results for changes in the population structure of rare and common, increasing and declining species in both habitats are summarized in Fig. 6.

Figure 5.

The trend in the skew of the distribution of densities within each species over time, plotted against the species’ mean density over all years, for (a) farmland and (b) woodland species. Filled symbols and solid lines are declining species, open symbols and dashed lines are increasing species. Fitted lines are from a General Linear Model modelling the trend in skew as a function of ln(mean density) and occupancy trend, and the interaction between density and trend (see Table 3). The dotted horizontal line in each case is at 0 (no trend).

Table 3.  Slopes of the relationship between ln(mean density) and trend in skew in density for declining and increasing farmland and woodland species
 Declining speciesIncreasing speciesDifference in slopes
Farmland280·16 ± 0·89536−0·12 ± 0·1162·380·0206
Woodland260·31 ± 0·10525−0·15 ± 0·1293·560·0009
Figure 6.

Mean skew in the distribution of population densities in (a) farmland and (b) woodland birds. For each species, the trend in skew over time was estimated using a linear model; we then used this fitted model to estimate skew in this species at the beginning and end of the study period. Here we plot mean start (open bars) and end (filled bars) values of skew separately for increasing and declining, rare and common species. Rare species are defined as those with a time-averaged mean density falling within the first quartile of the interspecific range; common species as those with a mean density in the top quartile of the interspecific range. Negative values indicate a left-skewed distribution (long left tail, i.e. most populations are large), positive values a right-skewed distribution (long right tail, i.e. most populations are small).


interspecific patterns

We have documented a significant decline in the interspecific abundance–occupancy correlation for British farmland and woodland birds. In addition, nonparametric smooth functions revealed that this decline in correlation has been accompanied on farmland by a decrease in the degree to which the relationship is curved. Given that the dynamics of the farmland bird group we have studied have been to a large extent driven by anthropogenic effects (e.g. Chamberlain et al. 2000), and that there are emerging links between changes in woodland habitats and woodland bird populations (Amar et al. 2006), this analysis adds to the growing consensus that human-driven habitat change impacts macroecological patterns (Gaston 2004).

Fisher & Frank (2004) documented a strong temporal trend in the slope of the linear relationship of abundance on occupancy of marine fish in eastern Canada, with a weaker (but still significant) positive trend in the correlation. From Fisher & Frank's fig. 6(b), a decrease in variance in both abundance and occupancy seems likely, as abundant widespread species have been directly targeted by commercial fisheries while some rarer species have become more common. This contrasts strongly with the situation in woodland birds, where the decrease in the interspecific correlation has been driven primarily by an increase in interspecific variance in occupancy. Here, common species have become more common while rare species have become rarer, in line with the ‘biotic homogenization’ scenario (e.g. McKinney & Lockwood 1999), in which ‘winners’ (abundant, generalist species) systematically replace ‘losers’ (rare, specialist species), although additional analyses would be required to properly test this hypothesis. In farmland birds, the observed patterns appear to derive from a decline in the covariance of density and occupancy over time. Thus, the drivers behind changes in the interspecific abundance–occupancy correlations in marine fish, woodland and farmland birds appear to be rather different; but the trend in interspecific correlation may serve as a useful indicator that some significant macroecological change has occurred.

The data set that we used contained detailed surveys of bird species in specific habitat types, known to be used by birds. This may not be the case in other situations, for example fisheries surveys. Thus, while several bird species were recorded at or near full occupancy in every year, the most frequently sampled species of fish in Fisher & Frank's (2004) study occurred in only 71% of samples. However, for the farmland species that we considered, occupancy on farmland was tightly correlated with occupancy calculated across all CBC plots, regardless of habitat (r = 0·93), and also with the number of 10 km squares occupied throughout Britain (from Gibbons et al. 1993) (r = 0·77). Estimating occupancy at this largest feasible scale, the interspecific relationship remains saturating, and all seven species that have an occupancy (averaged over all years) on farmland > 0·9, also occupy > 90% of the sampled 10 km squares. Thus, although our density measures are habitat-specific, occupancy (for these common breeding birds at least) remains remarkably consistent across habitats and scales.

Fisher & Frank (2004) were able to provide a convincing causal link between direct exploitation by a commercial fishery, changes in abundance and distribution, and changes in the abundance–occupancy relationship. It is harder to establish a causal relationship with the farmland, and particularly the woodland birds.

That said, on farmland we know that the patterns we report occurred over a period of intensification of agriculture (Chamberlain et al. 2000), which has led to changes in habitat quality for many farmland bird species (Robinson & Sutherland 2002); and that farmland bird densities respond directly to changes in habitat quality (Whittingham et al. 2005). However, many aspects of habitat quality have been affected by intensification, and these are frequently species-specific. For example, winter food supplies (Robinson & Sutherland 1999), summer availability of insect prey (Potts & Aebischer 1991), and the availability of suitable nesting sites (Green & Stowe 1993) have been altered in a species-specific manner, leading to variable responses across species.

The situation is less clear in woodlands, and hypotheses linking habitat change in woodlands to woodland bird populations remain tentative at present (Amar et al. 2006). We note, however, that year-to-year differences in mean density are generally not strongly correlated between woodland and farmland habitats for species that occur in both (mean ± SE correlation between density changes across habitats = 0·17 ± 0·039, n = 51 species; each correlation includes 31 year-to-year differences in mean density on farmland and woodland). This strongly implies that habitat-specific factors (primarily anthropogenic in origin) are driving the patterns we observe, rather than more general processes such as climatic effects on environmental productivity.

Finally, it would be interesting to continue monitoring the abundance–occupancy relationship in the same bird species to see whether an equilibrium is reached, for instance if the densities of remaining low-density species are depressed over their ranges, consistent with the lagged responses of density and occupancy observed in other taxa (Conrad, Perry & Woiwod 2001). This would be particularly interesting in the woodland species, where it is changes in occupancy, rather than abundance, which appear to be driving the interspecific trends. If densities respond to these occupancy changes this would run counter to the general consensus that it is changes in density that drive changes in occupancy (Gaston & Blackburn 2003b), although it would be consistent with various habitat fragmentation and metapopulation scenarios, whereby population extinction precedes density changes in remaining populations. Alternatively, remediation efforts might be expected to reverse the observed trends, and monitoring the abundance–occupancy relationship may provide a relatively simple method to assess the effectiveness of the many schemes now being introduced at a national and European level designed to improve habitat for farmland birds (see Greenwood 2003 for examples).

intraspecific patterns

The general trend in the interspecific relationship results from temporal dynamics of the abundances and occupancy of individual species. The intraspecific relationships of relatively rare species generally depart further from the interspecific relationship compared with those of relatively common species (Fig. 4c,d), a phenomenon predicted by models associating density and occupancy with the distribution of habitat suitability (Freckleton et al. 2006). In addition, our results show rare species consistently have abundance–occupancy slopes that are shallower than expected. Statistical constraints imposed by the asymptotic nature of the interspecific relationship might be expected to produce higher variation in the intraspecific relationships of rare compared with common species, and increases in measurement error (for instance, reduced detection probability) in rare species might also produce such an effect. However, such constraints would not predict the directional increase in variance that is observed.

Increasing and declining species also differ in their relationship between density and relative intraspecific abundance–occupancy slope (Fig. 5). The trend for rarer species to depart further from the interspecific relationship than common species is stronger in declining than in increasing species, at least for woodland birds; and the intraspecific relationships of increasing species generally lie closer to the interspecific relationship at any given density in both habitats.

Our results are in broad agreement with predictions from a previous study (Freckleton et al. 2006) regarding correlated changes in occupancy and the skew of densities within species. Thus, common species generally have left-skewed distributions that change predictably in both increasing (becoming more skewed) and declining (becoming less skewed) species (Fig. 6). Rare species have right-skewed distributions that change in a less-predictable fashion (Fig. 6). These results have intuitively appealing explanations – for instance, population increases at occupied sites together with the establishment of new, small populations will increase the left-skew of common, increasing species; the founding of new, small populations will increase the right-skew of rare, increasing species – and provide new evidence supporting strong linkages between density, occupancy, the distribution of population sizes across sites, and the distribution of suitable habitat.


Our findings add to the growing consensus that macroecological patterns change over human time-scales (Murray & Dickman 2000; Gaston & Blackburn 2003a; Fisher & Frank 2004; Gaston 2004; Jennings & Blanchard 2004). In addition to frequency distributions of individual variables, we confirm that such changes are manifested in an important bivariate relationship between macroecological variables, the abundance–occupancy relationship (see also Fisher & Frank 2004). Indeed, examining individual variables may prove misleading. For instance, low density species, and particularly those experiencing long-term declines, tend to show unusually weak intraspecific abundance–occupancy relationships, suggesting that continued declines in occupancy may not be detected if local population density alone is monitored. Hartley & Kunin (2003) showed that detecting declines in distribution is dependent on the resolution at which occupancy is measured (where population size is the highest resolution possible). The correlation between abundance and occupancy may provide a simple means to overcome such problems, and its temporal trend could present a more accurate picture of community change than changes in either variable alone. The differences between habitats for bird species, and between birds and marine fish (Fisher & Frank 2004), suggest that more work is required before an observed change in the interspecific abundance–occupancy relationship can be related directly to degradation or recovery of a community. However, temporal trends in the abundance–occupancy relationship may be used as an alert that some form of community change is occurring. This highlights the potential to use macroecological patterns as an indication of the state of ecosystems, and the extent to which they have departed from an undisturbed state (Jennings & Blanchard 2004).


Thanks to Alison Holt and members of the York University Environment Department Post-doctoral Research Seminar series for comments and discussion. David Storch and two anonymous reviewers provided very useful comments. This research was funded by the NERC (grant NER/M/S/2003/00096 to RPF). RPF is a Royal Society University Research Fellow.