Bird diversity and environmental gradients in Britain: a test of the species–energy hypothesis


J. J. Lennon (tel. + 44 (0)113 2332825; fax + 44 (0)113 2333835; e-mail


1. We tested the species diversity–energy hypothesis using the British bird fauna. This predicts that temperature patterns should match diversity patterns. We also tested the hypothesis that the mechanism operates directly through effects of temperature on thermoregulatory loads; this further predicts that seasonal changes in temperature cause matching changes in patterns of diversity, and that species’ body mass is influential.

2. We defined four assemblages using migration status (residents or visitors) and season (summer or winter distribution). Records of species’ presence/absence in a total of 2362, 10 × 10-km, quadrats covering most of Britain were used, together with a wide selection of habitat, topographic and seasonal climatic data.

3. We fitted a logistic regression model to each species’ distribution using the environmental data. We then combined these individual species models mathematically to form a diversity model. Analysis of this composite model revealed that summer temperature was the factor most strongly associated with diversity.

4. Although the species–energy hypothesis was supported, the direct mechanism, predicting an important role for body mass and matching seasonal patterns of change between diversity and temperature, was not supported.

5. However, summer temperature is the best overall explanation for bird diversity patterns in Britain. It is a better predictor of winter diversity than winter temperature. Winter diversity is predicted more precisely from environmental factors than summer diversity.

6. Climate change is likely to influence the diversity of different areas to different extents; for resident species, low diversity areas may respond more strongly as climate change progresses. For winter visitors, higher diversity areas may respond more strongly, while summer visitors are approximately neutral.


Geographical gradients in species diversity are almost universal across taxa. Understanding this is no longer an academic matter: given the current rapidity of global climate and land use change, we urgently need to find a working understanding of what controls diversity (Lawton et al. 1996). Despite the obviousness of patterns in diversity, this is not easy. Many explanations have been suggested over the years for diversity patterns (reviewed by Ricklefs 1987; Rohde 1992; Rosenzweig 1995; Turner, Lennon & Greenwood 1996) and it is questionable if these have lead to understanding or confusion: most of them are very difficult to test. The explanation we consider in this paper, namely the species diversity–energy hypothesis, is something of an exception, as the predictions are relatively clear and the necessary data are available: the strongest determinant of diversity should be some measure of energy.

The species–energy hypothesis

Perhaps for these reasons of relative simplicity, the species diversity–energy hypothesis has attracted much interest (Wright 1983; Currie & Paquin 1987; Turner, Gatehouse & Corey 1987; Turner, Lennon & Lawrenson 1988; Adams & Woodward 1989; Currie 1991; Currie & Fritz 1993; Wylie & Currie 1993; Wright, Currie & Maurer 1994; Fraser & Currie 1996; Turner et al. 1996). Although it has found a measure of support, there is some evidence against it, particularly as an explanation for continental-scale trends in tree diversity, where long-term historical factors may be paramount (Latham & Ricklefs 1993; McGlone 1996).

The main prediction is that local species diversity increases with energy availability. This ‘energy availability’ is usually taken (rather crudely) to mean average temperature. For example, the latitudinal temperature gradient is suggested to be the cause of the latitudinal diversity gradient shown by most taxonomic groups. But even if there is a causal relationship between climate and diversity, it is not known how this works. In the diversity literature, we can identify two broad categories of possible mechanisms: the indirect and the direct. In the indirect route, temperature influences diversity through complex pathways involving effects on resource levels, population densities, competition, predation and other biotic interactions. Most potential mechanisms fall into this indirect category. For example, temperature may control the quantity and seasonal availability of food and so influence species population dynamics (this may in turn produce cascading changes in the outcome of species’ interactions). These net effects on population dynamics may alter local extinction risk, and so temperature may indirectly influence local species diversity. In contrast, a single direct mechanism has been proposed (Turner et al. 1988, 1996), in which the temperature directly experienced by individuals is of fundamental importance. Here, diversity is controlled by the relatively simple effects of temperature on the energy budgets of homeotherms; these effects on individuals determine local population densities and hence control local extinction probabilities. Because cold locations impose greater thermoregulatory loads, an individual has to devote relatively more energy to thermoregulation. As a result, less energy is available for growth and reproduction, so colder areas will tend to have lower population densities than warmer areas. This lower density will tend to make local populations more susceptible to extinction. Spatial variation in diversity therefore occurs as a consequence of climatically driven variation in local extinction probability.

Testing the hypothesis using british birds

Because temperature and latitude are correlated, it can be difficult to disentangle the effect of temperature from that of latitude. However, temperature is predicted to be more strongly associated with diversity than latitude alone. While earlier workers have relied heavily on this prediction, the task is complicated by the fact that most climatic and environmental variables are correlated, and this correlation structure can make it challenging to disentangle the causal factors.

Because many species of British birds are migratory, and additionally some species move within the country between the seasons, the spatial pattern of diversity changes in time. This opens up the possibility of seeing if diversity tracks seasonal change in climate. Usefully, oceanic influences mean that the spatial distribution of climatic factors in Britain changes dramatically from summer to winter. Summer is dominated by a south to north decline in temperature, but in winter the gradient swings round to a south-west to north-east direction. We can use this ‘natural experiment’ to predict that the changing pattern of climate between seasons should change the spatial pattern of diversity, if there is a direct link between energy input and diversity.

In earlier work, we found some associations between climate and bird diversity (Turner et al. 1988; Lennon 1990; Turner et al. 1996). Here we update and improve upon this work by using more recent bird survey data, better climatic data, most of the British bird fauna and a much larger number of quadrats. We simultaneously assess the effects of climate, habitat (as land use), topography and geographical location. We also introduce analytical methods that have several advantages over those usually used.


We took an individual species approach. For each species’ spatial distribution, we fitted a logistic regression model using climatic, habitat, topographic and spatial location variables. These individual species models were then amalgamated to give a composite species diversity model: this describes diversity as a function of the environmental factors. We defined four bird assemblages based on the season of the year (winter or summer) and migration status (resident or visitor).

This individual species approach has several advantages over modelling diversity directly: (i) it allows estimation of the contribution of each species to diversity–environment associations; (ii) it also allows detection of trends associated with individual species’ characteristics, such as body mass; and (iii) diversity is not constrained to respond to the environment linearly; this is achieved with relative simplicity. For comparative purposes, we also applied the more limited approach of ordinary multiple regression, where species diversity is modelled directly, and we give the simple pairwise correlations between diversity and environmental factors.

Predictions of the species–energy hypothesis

For the species–energy hypothesis to be rejected, the association between diversity and temperature should be weaker than its association with other environmental factors. We used a broad set of competing environmental variables: precipitation, habitat type (and diversity), spatial location (latitude, longitude, altitude and distance from the coast) and topographic variation. These are obviously not independent of each other: habitat diversity is a function of topographic variation, while topographic variation is not randomly placed within Britain. Distance from the coast can also be interpreted as proximity to an extreme example of habitat variation.

For both the direct and indirect mechanisms, species’ interactions (between bird species and between birds and other species) are predicted to be less important than temperature. Although the indirect mechanism may include interactions, these are predicted to be overwhelmed by the effects of temperature on diversity.

To support the direct mechanism, diversity should be determined by directly acting climate (Table 1); this means that the summer patterns of diversity should match summer temperature, and the winter diversity patterns should match winter temperature. Diversity of resident species (those present all year) should increase with higher summer and winter temperature. In contrast, diversity of visiting species should be influenced only by the season in which they are present: summer visitors by summer temperature, winter visitors by winter temperature. However, as residents are present throughout the year, winter temperature should influence summer resident diversity and, similarly, summer temperature should influence winter resident diversity, but these cross-season effects should be weaker than the effect of the same season. Because the resident birds are present the year round, they are less useful discriminants between causes than the visitors.

Table 1.  The predicted relationships, according to the direct and indirect mechanisms, between diversity and temperature in summer (S) and winter (W) for resident species in summer (SR), visitors in summer (SV), residents in winter (WR) and visitors in winter (WV). All associations are predicted to be positive. + += strong association, += moderate association, 0 = weak or no association
SR+ +++ +0
SV+ +0+ +0
WR++ ++ +0
WV0+ ++ +0

The direct mechanism makes a further prediction about the relationship of diversity and body mass (Cousins 1989). Lighter birds should show a stronger relationship with temperature, because they have to devote more energy per unit mass to thermoregulation. Because of this (all else being equal) lighter species should contribute more to the match between diversity and temperature. It has been brought to our attention (John Lawton, personal communication) that this test is vulnerable to the net outcome of the opposing effects of larger body mass on local extinction risk (smaller annual population fluctuations = lower extinction risk, but lower densities = higher extinction risk). We do not know if these two forces are equal and opposite. It is possible that the net outcome could cancel an effect of energetic factors working through the direct mechanism or, alternatively, a trend found for body mass might be a consequence of the population fluctuation/density trade-off.

The indirect mechanism, perhaps operating via productivity (which depends mainly on summer rather than winter temperature) predicts that diversity in both summer and winter is determined mainly by summer temperature alone (O’Brien 1998).

All relationships between diversity and temperature are predicted to be positive by both the direct and indirect hypotheses.

Bird distributions

The spatial distribution data consisted of records of species’ presence/absence in 2362, 10 × 10-km, quadrats covering the British mainland and most of the islands. Ireland, the Isle of Man and the whole of the northern isles were excluded, as were coastal quadrats containing less than 50% land. The species’ distributions result from two distinct surveys co-ordinated by the British Trust for Ornithology: a winter survey in the years 1981–83 (Lack 1986) and a summer survey during 1988–91 (Gibbons, Reid & Chapman 1993).

We retained the introduced species in the analysis. While potentially there may be problems if current distributions still reflect their original point of release or escape, the fact that some introduced species (e.g. Athene noctua Brehm, little owl) are now well established means that their presence has some ecological meaning. However, we did exclude the marine species. These depend on an environment that differs greatly from the terrestrial environment. Deciding on which species these are was not straightforward. There is a spectrum from strictly oceanic seabirds, e.g. Fulmarus glacialis L. (fulmar), to species whose distributions are mainly terrestrial but have maritime components, e.g. Larus ridibundus L. (black-headed gull); some use both land and sea, switching affinities between seasons, e.g. Larus argentatus Pontopp. (herring gull). Where what is usually considered to be a seabird has a substantial inland distribution in one of the two seasons, we included it in the analysis for that season. ‘Erratics’ were also excluded; these are typically species recorded from a few individual birds in a handful of quadrats. However, we did not exclude species solely on the basis of coverage. For example, we considered very small remnants left behind by a large migrating population, e.g. Apus apus L. (swift), to be less ecologically meaningful than a few records for a species that is merely geographically very rare because it is on the cusp of regional extinction or recolonization, e.g. Grus grus L. (crane).

Bird body masses (averaged across sexes, for adults in the breeding season) were obtained from the literature (mainly Hickling 1983 and Cramp 1977) and are tabulated in Lennon (1990).

Season/migration assemblage classification

We categorized species into four migration/seasonal assemblages: residents in their summer distributions (SR, 120 species), residents in their winter distributions (WR, 102 species), visitors in the summer (SV, 78 species), and visitors in the winter (WV, 69 species). These assemblages overlap in species composition. Usually, a summer resident is also a winter resident (as implied by ‘resident’). However, when a species has substantial internal migration within Britain, deserting breeding sites for wintering sites for example, we departed from the usual ornithological meaning of ‘visitor’ and ‘resident’ and classified such species as visitors in both their summer and winter distributions, i.e. as both SV and WV, e.g. Regulus ignicapillus Temm. (firecrest). This classification was also applied to species that are visitors from overseas in both seasons, e.g. Sylvia atricapilla L. (blackcap). Similarly, where a species is mainly a visitor to Britain but leaves a remnant behind in the other season, we treated the species as a resident in this other season, e.g. Circus aeruginosus L. (marsh harrier) is both SV and WR. We did not classify a species as both a resident and a visitor within a single season. The species are listed in Lennon, Greenwood & Turner (2000).

Environmental data


Two climatic factors for each of summer and winter were used: monthly mean temperature, and monthly mean precipitation (Fig. 1). Summer was defined as May, June and July, and winter as December, January and February. These data were derived from meteorological recording station readings for the period 1961–90 using surface interpolation techniques (Barrow, Hulme & Jiang 1993). The monthly averages were converted to winter and summer seasonal averages (corrected for month length).

Figure 1.

Climatic variables: spatial distribution of mean temperature (Celsius) and mean monthly precipitation (mm) in summer (May–July) and winter (December–February). The temperatures are not corrected to sea level. The marginal numbers are National Grid two-figure northings and eastings.

Spatial coordinates

We used quadrat centre latitude (as National Grid northing), longitude (as easting) and mean altitude within the quadrat as explanatory variables. Distance from the coast (log transformed) was included to account for maritime influences. The position of the coast was defined as the midpoint between the high and low spring tides.


For each quadrat, four functions of altitude were calculated from a digital elevation model. These were the mean, maximum, minimum and standard deviation of 400 evenly spaced altitudes (at 500-m latitude and longitude intervals) within the quadrat. For maps of the mean and standard deviation functions of altitude see Lennon & Turner (1995). Topographic variation indicates the fragmentation of altitude; bird species with preferences for particular altitudes are likely to perceive topographic variation as fragmented habitat.


We used the land use classification of Fuller, Groom & Jones (1994) based on remote sensing. We selected seven of the 25 cover types as likely to be influential, if not for species diversity at least for particular species (Table 2). We amalgamated some of the basic cover types; coniferous and deciduous woodland were merged, as were heath and moorland grass. We took the seven habitat coverages and calculated an index of habitat diversity using the Shannon–Wiener information index (Ludwig & Reynolds 1988).

Table 2.  Habitat principal components (PC) defined as weighted contributions of seven land use coverages within each quadrat. These three (of the maximum of seven) PC summarized approximately two-thirds of the spatial variation in these habitat types. Each habitat type is given in terms of its Fuller et al. (1994) land cover classification index
hab1hab2hab3Contributing variableFuller cover class
−0·15−0·24+0·79Woodland (deciduous and coniferous)15 +16
−0·37+0·36−0·31Bog (upland and lowland)17 +24
+0·51+0·40+0·16Suburban/rural development20
−0·18+0·36−0·17Inland water2
−0·44+0·26+0·07Heath and moorland grass5 +9
+0·44−0·33−0·39Tilled land (including arable crops)18
 301915Percentage of variation in habitat type explained by each PC 

Statistical methods

We introduce a novel composite logistic regression method, and for comparison the more familiar ordinary linear multiple regression and pairwise correlation.

Principal components – summarizing habitat and topography

To reduce the number of variables in the analysis, we calculated the principal component (PC) scores of the topographic variables and, separately, the PC of the habitat coverages (Table 2). This allowed us to simplify the analysis but still detect the effects of habitat or topography. The first topographic PC explained 82% of variation, and was a roughly equal weighting of all four variables. The remaining three topographic PC were discarded from further analyses. The first habitat PC was weighted heavily towards urban and suburban areas, and agricultural land, but against woodland, heath, bog, moorland and freshwater. The second PC again indicated urban/suburban areas, but also heath, bog, moorland and freshwater. The third PC weighted towards urban/suburban as well, but more weakly, and indicated an absence of agricultural land but the presence of woodland. The remaining four habitat PC were discarded.

Standardization of variables

All variables, with the exception of observed species diversity values, were standardized to a mean of zero and unit variance. This simplified comparison of the effects of different environmental factors, as they varied widely in measurement scales.

Ordinary linear regression

For the ordinary linear regression (OLR), the diversity of each assemblage was fitted to the familiar model:


where n is the number of environmental variables (in this case n = 13), xk is the kth environmental variable and

inline image is its estimated regression coefficient.

A ‘stepwise’ independent variable selection algorithm was used (SAS Institute 1990): the model finally fitted contained a subset of the independent variables that were selected according to significance criteria. Using a Bonferroni correction for multiple tests (Sokal & Rohlf 1995), both the entry and removal significance criteria were set at a conservative P < (0·05/13), as there were 13 independent variables (Table 3), assuming an α of 0·05 per variable.

Table 3.  Environmental variables
longLongitude, as National Grid northing
latLatitude, as National Grid easting
altMean altitude
topTopography, first PC score
dseaLogarithm of shortest distance to coast from quadrat centre
stempSummer temperature, mean of May, June and July
srainSummer precipitation, mean of May, June and July
wtempWinter temperature, mean of December, January and February
wrainWinter precipitation, mean of December, January and February
hab1Habitat, first PC score. See Table 2
hab2Habitat, second PC score. See Table 2
hab3Habitat, third PC score. See Table 2
hdivHabitat diversity as Shannon–Wiener index. Using habitats in Table 2

Composite logistic regression modelling of species diversity

The ordinary multiple regression model implies that the response of diversity to a particular environmental factor is the same across all quadrats. This is an undesirable constraint. For example, equation 1 shows that a quadrat that differs by one degree of temperature from its otherwise identical neighbouring quadrat is assumed to have a diversity that is a fixed number of species different from these neighbours, irrespective of whether the quadrat has 10 species or 50 species. The composite logistic regression (CLR) approach we now describe is not constrained in this way. For example, a quadrat that differs by one degree from its neighbour may have an additional number of species that varies according to the particular set of species present in the quadrat, i.e. the identity of species matters.

Using logistic regression, the estimated probability inline image of finding a particular species present in a quadrat that has a set of n environmental variables {x1, x2xn} was modelled as:


In our analysis, again n = 13. The environmental variables xk are identical to those used in equation 1, but the inline image and inline image coefficients are quite different. A positive inline image coefficient indicates that an increase in xk makes the presence of the species more likely. As in the case of the OLR analysis, each model was fitted using the stepwise algorithm (SAS Institute 1990). The same significance criteria for variable introduction/removal were applied.

Combining the individual species models to estimate diversity

The models for each species were combined to give a diversity model for each assemblage. Quadrat diversity was estimated as the sum of the probability of finding each species in the quadrat:


where inline image is estimated diversity, S is the number of species in the assemblage, and the inline image are the probabilities given by equation 2.

This quadrat diversity estimate has two kinds of variance connected with it: that produced by the uncertainty associated with the estimation of the inline image coefficients, and that implied by the fact that what is estimated is itself a sum of probabilities. For the purposes of our analyses we ignored both these sources of variation and take the inline image values as given with uniform accuracy across species.

There is no difference between the CLR and the OLR in their susceptibility to any difficulties that might arise from species’ interactions.

Assessing the importance of each environmental variable

It is not easy to see directly from equation 3 how the environmental factors affect diversity: their relative influence is obscured. However, we can still quantify the sensitivity of diversity to each environmental variable. Imagine changing one environmental factor by a small, standardized, amount while holding the other factors constant; when this is done for each factor in turn, the factor that produces the largest change in diversity is the most influential. More formally, this is the rate of change of diversity with respect to the kth environmental variable (the partial differential coefficient). This is found by differentiating equation 3 with respect to xk. The resulting equation (equation 4) is relatively simple: it expresses the sensitivity (or response) of diversity to an environmental factor as the sum of the sensitivities of the individual species to this factor. In a particular quadrat, the response of a species to the kth environmental factor depends on the both the probability of presence and the weighting given to the kth factor (the βk coefficient):


where inline image . The overall response of diversity (across all quadrats) is the sum of these quadrat responses. Species with opposite signs of response (the signs of their βs) tend to cancel each other out. A large absolute value for a particular environmental factor indicates greater influence.

The effect of body mass on bird diversity–environment associations

This reductionist way of modelling diversity, because it uses individual species’ sensitivities to environmental factors, allowed us to quantify the contribution of each species to the relationship between diversity and each environmental factor. In particular, we looked for any tendency of lighter species to contribute more to the sensitivity of diversity to temperature. The overall contribution from a single species is the inline image component of equation 4 (summed across all quadrats). If lighter species contribute more strongly to the diversity-temperature relationship, as is predicted by the direct hypothesis, we would expect to see a negative correlation between these contributions and species mass.

Significance testing and spatial autocorrelation

Both the species distributions and the environmental variables are spatially autocorrelated. Conventional significance testing is of doubtful validity here, because the assumption of independence of the error term between measurements is violated. This leads to overestimation of the true degrees of freedom (Haining 1990; Borcard, Legendre & Drapeau 1992). This problem is usually ignored in the geographical ecology literature (Turner et al. 1988, 1996; Currie 1991; Newton & Dale 1996). Although corrected tests are available for the special cases of correlation of two normally distributed variables (Clifford, Richardson & Hemon 1989) or two binary variables (Cerioli 1997), a coherent approach to significance testing for multivariate spatial processes has yet to emerge. Therefore, we do not report misleadingly precise significance levels for the associations we describe below.


Spatial patterns of diversity

The diversity patterns (Fig. 2) were all positively correlated with each other (Table 4). Summer and winter residents were strongly correlated, as expected, but the correlation between winter residents and visitors was strongest. Visual inspection suggested that the latitudinal gradient in diversity was most pronounced in the resident distributions; the winter visitor distribution suggested combined coastal, altitude and latitude effects, while the pattern for the summer visitors suggested, if anything, a reversed latitudinal diversity gradient (confirmed by correlation; Table 4). Because of adverse weather, coverage in the winter distribution survey may have been less comprehensive: most or all of the few scattered low winter diversity quadrats (less than 1%) were probably under-recorded. Species diversity (richness) is not as easy to measure as its simplicity might suggest (Gaston 1996) and sampling effort biases, if present on geographical scales, can cause serious problems. That there was no marked bias in our analysis was supported by the mismatch between the recorder effort maps and the diversity maps in the distribution atlases (Lack 1986; Gibbons et al. 1993).

Figure 2.

Diversity patterns of the four bird assemblages, as the number of species present in each 10 × 10-km quadrat.

Table 4.  Pairwise correlations (Pearson) between the species richness (as in Fig. 2) of different assemblages and between species richness and the environmental variables. The four strongest correlations with environmental variables are indicated for each assemblage. n = 2362 throughout
long0·480·05−0·66 (2)0·24
lat−0·50 (2)0·15 (1)0·56−0·25
alt−0·420·03−0·59−0·53 (2)
top−0·430·05−0·62 (3)−0·55 (1)
stemp0·59 (1)−0·050·77 (1)0·44 (4)
srain−0·49 (3)0·02−0·62 (3)−0·37
wtemp0·33−0·13 (2)0·530·48 (3)
wrain−0·49 (3)−0·01−0·62 (3)−0·38
hab10·39−0·12 (3)0·580·33
hab30·210·09 (4)0·150·00

Plots of diversity against temperature

Scatter plots of diversity vs. summer temperature generally showed strong positive relationships (Fig. 3; correlation coefficients in Table 4). While the relationship for SV was very weak, the diversity of the resident and winter visitor assemblages was clearly associated with summer temperature. There was also a suggestion that resident diversity asymptotes at higher temperatures, and that the WV have greater variance in diversity at higher temperatures. While spatial autocorrelation makes interpretation difficult, and it may just be a simple mean–variance trend, there is a suggestion for WV that points are restricted to a roughly triangular area in diversity–temperature space.

Figure 3.

Diversity plotted against summer temperature (Celsius). There is a clear relationship with summer temperature for all but the summer visitor assemblage.

Pairwise correlations between diversity and environmental factors

Summer temperature was the strongest correlate for both resident assemblages; second placed factors were latitude (summer) and longitude (winter), both negative. For SV, latitude has the strongest association, but is weak in absolute terms. For WV, topographic variation was strongest (negative) and altitude second, with summer temperature ranked only as fourth behind winter temperature (Table 4). In summary, using pairwise correlation, across all assemblages summer temperature was most influential, with simple geographical trends (latitude, longitude and altitude) the second most influential factors.

Ordinary linear multiple regression

The regression coefficients for each environmental variable included in the final regression model for each assemblage are given in Table 5. This analysis suggested that summer temperature was the most influential factor for all but the winter visitor assemblage. WV had summer temperature as the second most important, after distance from the coast dsea (these results for WV are in contrast with the correlations, which ranked summer temperature in fourth place and dsea in joint sixth place). The next most important factor, taking all assemblages together, was the negative effect of winter temperature (pairwise correlation suggested lat, long and alt). Habitat effects were: habitat diversity is consistently positive; hab3 has three positives; and hab1 is negative (rank 4) for both summer assemblages. Summer visitors show a noteworthy rank-2 effect of latitude. The amount of total variation in diversity explained by the model is high for residents in winter, lower in the summer, and lower for visitors than residents.

Table 5.  Ordinary linear regression of diversity on environmental variables, showing the (standardized) b coefficients from equation 1. See Table 3 for definition of environmental variables. Models were fitted using stepwise variable selection. The four strongest associations are ranked by magnitude for each assemblage
model r20·460·120·650·41
lat+2·73 (2)
dsea−5·03 (1)
top−2·92 (4)
stemp+9·78 (1)+4·68 (1)+12·8 (1)+4·29 (2)
srain−2·86 (3)−1·46
wtemp−4·85 (2)−1·94 (3)−2·15 (4)−3·13 (3)
wrain−1·84−5·77 (2)
hab1−2·15 (4)−1·77 (4)
hab3+2·27 (3)+0·78+1·82

Using pairwise correlation led to some other areas of disagreement with OLR, e.g. winter temperature was negatively associated with all four assemblages using OLR but the simple correlations were positive for all but the summer visitors.

Individual species models

Logistic regression fitted almost all species’ distributions with nominally significant environmental variables: the exceptions were one winter visitor, four summer residents and six summer visitors. All these species were present in less than 1% of quadrats; their presence or absence made little difference to diversity patterns. Surprisingly, as distributions of extreme commonness have as little information as those of extreme rarity, no common species generated nominally insignificant models. This may be partly explained by asymmetry in the number of common vs. rare species: taking 5% and 95% coverage as arbitrary limits of rarity and commonness, the number of species rare and common for each of the assemblages was 27,10 (SR), 27,2 (SV), 15,3 (WR) and 13,0 (WV), respectively. If this asymmetry does not explain fully the absence of common ‘intercept only’ species in the results, then absences from small areas in near-ubiquitous distributions of common species could be more environmentally meaningful than a small number of presences for rare species.

Across the four assemblages, there were strong trends in the influence of particular environmental variables for individual species, as identified by the magnitudes of the inline image coefficients (Fig. 4). Coefficients usually negative (indicating an association with species’ absence) were altitude and winter precipitation. Those usually positive (associated with species’ presence) were summer temperature, longitude, hab3 and habitat diversity. On average, the single most influential environmental factor for individual species’ distributions was summer temperature.

Figure 4.

The individual species–environment associations. Frequency distributions of individual species’ logistic regression β coefficients (equation 2) within each of the four assemblages for all environmental variables. The shaded bar represents the interquartile range and the median is marked within this; the line extensions from each box are the largest and smallest values, excluding outliers (points more than 1·5 times the interquartile range distant from the box edges). The asymmetrical position of the median within most interquartile ranges indicates that most of the distributions are skewed.

Composite species diversity models

Relative importance of environmental factors

The CLR analysis found that summer temperature was the strongest determinant of diversity. For all four assemblages, summer temperature showed the highest median response (Fig. 5). Only for the winter visitors did other environmental variables even approach comparable strength (distance from coast and topography: pairwise correlation suggested altitude and topography, while OLR suggested distance from coast, winter temperature and topography).

Figure 5.

The diversity–environment associations. The response of the four diversity distributions to each environmental factor according to the composite logistic regression model. Each bar represents the responses in each of the 2362 quadrats for each environmental variable (see Fig. 4 for meaning of bar symbols). For example, the bars for summer temperature show the frequency distribution of the responses mapped in Fig. 6. The influence of each environmental variable is ranked by the distance of the median from the origin. The heavy dashed lines indicate the magnitude of the largest median response, which is summer temperature throughout. As there is no prior expectation of the sign of the richness–environment association for most variables, both the positive and the negative part of the response axis is marked with a dashed line.

The amount of variation in diversity accounted for by CLR was 51%, 11%, 69% and 47% for the SR, SV, WR and WV assemblages, respectively (calculated from the square of the correlation between the predicted and observed diversities). These figures are generally rather higher than those for the OLR models (Table 5) despite using identical data. Both models poorly predicted summer visitor diversity. The ranking (by magnitude) of environmental variables differed between the CLR and the OLR models, although the signs of the relationships were the same for all variables (with the minor exception of srain for the winter visitors).

Spatial trends in the sensitivity of diversity to environmental variation

Figure 5 describes the overall response of diversity to each environmental variable; this masks a large and often systematic variation in response across the country (Fig. 6). We might ask of these diversity–environment responses: does the response tend to be stronger in high diversity areas? This would be the case if the response grew with the number of species already present, indicating that a local increase in an environmental factor is ‘worth’ more extra species in a species-rich area. Visual inspection of the summer temperature response maps (Fig. 6) and the diversity maps (Fig. 2) suggested some trends; these were confirmed in scatter plots of response against diversity (Fig. 7). The correlation coefficients were −0·57, −0·09, −0·68 and +0·56 for SR, SV, WR and WV, respectively. On average across the whole study area, the residents tended to show elevated sensitivity to summer temperature in lower diversity areas; in contrast, the visitors were either approximately neutral (SV) or showed an elevated response in areas of high diversity (WV).

Figure 6.

Spatial patterns of sensitivity of diversity to summer temperature. The surface is the partial differential coefficient defined in equation 4; it shows how strongly diversity responds to a small uniform increase in summer temperature. To convert scale units to species per degree Celsius, multiply the standardized scale values by 1·55 (the standard deviation of summer temperature).

Figure 7.

The relationship between the response of diversity to a small increase in temperature (mapped in Fig. 6) and diversity (mapped in Fig. 2). The response tends to be larger in low diversity quadrats for resident species, but larger in high diversity quadrats for winter visitors. The summer visitors show no clear relationship.

Body mass and species’ contributions to the diversity–temperature association

The negative slope predicted by the direct mechanism was not found: smaller species were not more sensitive to temperature when other factors were accounted for. None of the rank correlations between individual species’ responses to summer temperature and body mass was significant (SR: Spearman θ = 0·16, P = 0·24, n = 54; SV: θ = −0·31, P = 0·37, n = 34; WR: θ = 0·17, P = 0·17, n = 67; WV: θ = −0·13, P = 0·51, n = 28, where n is the number of species), using each species’ median response (median of the inline image terms of eqn 4 across all quadrats).


Species–environment and diversity–environment associations

With increasing sophistication of methods (pairwise correlation, OLR, CLR), the importance of summer temperature for diversity becomes increasingly clear: summer temperature is the best explanation overall for British bird diversity.

Our results show that summer temperature explains the distribution of winter bird diversity better than winter temperature does. The only plausible ecological explanation seems to be that the summer temperature has a residual effect on species’ distributions in winter. If this is correct, then this seasonal lag or ‘productivity shadow’ is strong enough to overcome the direct effects of winter temperature, and is additional evidence that the species–energy mechanism operates indirectly (see the Introduction). This productivity shadow may take the form of summer accumulations of seeds, insects and other invertebrates, which remain available as food resources in the winter months (Turner et al. 1996).

Winter diversity of both residents and visitors is explained better than summer diversity by the set of environmental variables: the proportion of total variance accounted for is greater, by both the CLR and OLR methods. We can think of three (overlapping) explanations for this, although there are undoubtedly more: (i) the effective environmental factors for summer diversity have not been included; (ii) birds in their summer distributions perceive the abiotic environment in summer as more benign, so the distribution of each species and hence of species diversity is less constrained; (iii) birds are doing fundamentally different things in the summer and winter: summer is dominated by the need to nest and reproduce, while winter is more about feeding and surviving climatic extremes. Thus, the constraints on winter distributions may be simpler and more coherent, so allowing a better model fit using simple abiotic factors. In support of point (ii) there is much anecdotal evidence that winter is a particularly difficult time for birds (Moss 1995); this is also supported by Brown's (1995) findings of strongly individualistic dynamics in summer distributions of North American songbirds, contrasted with Root's (1988) evidence for a greater climatic control of bird distributions in the winter. Strictly, however, this latter is not comparing like with like, as Brown and his colleagues focused on population abundances in time and in different habitat strata, while Root concentrated on distributional boundaries. A synthesis of these approaches across both seasons is required.

The absence of trends associated with body mass in our results is perhaps surprising in view of the importance of body mass for features of life history, physiology and general ecology in many taxa (Schmidt-Nielsen 1984). However, it should be remembered that body mass is associated with population variability and density trends that act on local extinction risk in opposite directions (see the Methods). The large literature on species’ range size, body mass and abundance (Brown 1995; Gregory & Blackburn 1995; Hold et al. 1997) describes correlations between these three factors. From our results, these are not reflected in diversity–environment associations. This is, to some extent, consistent with the observation that body mass is a relatively poor predictor of abundance for British birds: life-history traits associated with fast offspring production explain more of the variation between species (Blackburn, Lawton & Gregory 1996). Although British birds show an unusual negative relationship between range size and body mass (Sutherland & Baillie 1993) this may be an artefact caused by Britain being too small, as larger-scale studies tend to show positive correlations (Gaston & Blackburn 1996). This suggests that associations between diversity and body mass cannot be ruled out entirely.

It is worth clarifying a potential confusion about how environmental constraints on individual species in turn control patterns of diversity. A strong or even a deterministic relationship between species’ distributions and their environment does not mean that there need be any relationship between diversity and environment. This arises, in part, because pairs of species may have opposite relationships with any given environmental gradient: an extreme example is a set of northerly distributed species balanced by a set of southerly distributed species, resulting in no net diversity gradient. This argument applies equally to other environmental gradients, no matter how complex, such as patterns of climate and habitat. However, if all or the majority of species are strongly positively associated with an environmental gradient, then it will necessarily be associated with diversity, all else being equal. Our results suggest that this is the case for British bird distributions when the gradient is summer temperature. The CLR analysis selects summer temperature as the most important factor for both individual species’ distributions and diversity patterns.

To estimate the effect of an environment factor on diversity, it is not enough to count the number of species having relationships with that factor: the factor that is associated with the greatest number of species is not necessarily the most influential. Both the strengths of the relationships for each species individually and their coverages (commonness versus rarity) must be considered. Thus, studies that merely count the number of species associated with various environmental factors (Tellería & Santos 1994) do not allow us to draw conclusions about the control of species richness. Similarly Root (1988) showed that more than half of North American bird species had a northern distributional boundary attributable to winter temperature (although in this case apparently reduced to sea-level, and see Repasky 1991)): this does not in itself show that winter temperature is the strongest influence on winter bird diversity in that region.

The contribution of environmental factors other than summer temperature

As an explanation of bird diversity patterns in Britain at the 10-km scale, the other environmental factors are less influential.

Other climatic factors

After taking into account all other variables, winter temperature is negatively associated with diversity across all four assemblages. It is not obvious why areas with colder winters should tend to be more diverse. It is possible that winter temperature is a surrogate for an unknown factor associated with reduced diversity, or that its negative weighting reflects an association with climatic continentality. Obviously, this merely shifts the problem of explaining the association, unless continentality can be shown to be more clearly associated with productivity or other factors influencing population abundances.

Winter precipitation also has a negative association with diversity across all four assemblages. Prolonged rainfall may reduce the insulation properties of birds’ plumage and thus demand increased energy expenditure, especially in winter. While this might help explain the effect on the winter assemblages (and absence of effect of summer rainfall on summer birds), clearly it does not explain the effect on summer birds.


The diversity of winter visitors is greater near the coast. This is clearly caused by the large numbers of wading and wetland birds that winter in coastal habitats (Fuller 1982). We can speculate that the generally negative effect of altitude (over and above the effects of climate, topography and habitat) partly reflects the low productivity of most British upland soils (McVean & Lockie 1969).


The associations between habitat and bird diversity are weaker than the link with temperature for all four migration/season assemblages; we should discount habitat as the main determinant of diversity patterns at this scale. This is not to say that habitat is of no consequence. Bird diversity of all four assemblages is positively associated with habitat diversity (perhaps more correctly called landscape diversity at this scale). Furthermore, the three habitat variables show moderate contributions to diversity in both seasons, particularly for the residents.

PC hab1, indicating urbanization and tilled land, apparently depresses diversity in summer. Such an effect of urbanization is not surprising; nor is an effect of tillage. Tilled land is intensively managed for production, and there is evidence that changes in the distributions of several bird species in Britain and mainland Europe are connected with agricultural activity (Baillie, Gregory & Siriwardena 1997). But why is there no effect of hab1 on winter birds? Possibly because urban gardens are important habitats for birds in winter (Moss & Cotteridge 1998).

Spatial scale

We only consider a single spatial scale (i.e. 10-km quadrat size). A detailed study of diversity–environment relationships at a range of scales would be particularly illuminating: scale is a neglected but potentially pivotal factor in spatial ecology (Levin 1992). There is weak evidence that habitat has more influence on bird diversity than temperature at finer scales, with the opposite perhaps the case at coarser scales (Böhning-Gaese 1997); although this involves the use of nested quadrats, such that the statistical tests are not independent between scales, this kind of approach is well worth pursuing.

Is there evidence for saturation in these results?

We have found that areas with high diversity of resident species are less sensitive to the most influential environmental factor (but note that the winter visitors reverse this relationship). Is there evidence here of local saturation of high diversity areas with resident species, in the sense of Cornell & Lawton (1992)? Perhaps, but to some extent this relies on viewing Britain as a closed system. Britain is a subset of the species’ geographical range for all of the birds we consider; this may have biased our analysis if Britain is an atypical part of their ranges. Britain is certainly atypical in that it has fewer species than parts of continental Europe at the same latitude (Gregory, Greenwood & Hagemeijer 1998). A repetition of our study at the European level would be worthwhile, especially as bird diversity is lower in the far south of the continent than further north (Gregory et al. 1998). Even if there is a saturation effect for the residents, why should diversity of winter visitors show the opposite trend, i.e. high diversity areas acquire more species than low diversity areas per degree of summer temperature? This is difficult, but perhaps one possibility is that the winter accumulation of species in coastal areas in large flocks is partly involved in this effect. This idea could be tested by using population abundances rather than just presence/absence data.

In any case, it is possible that colonization of Britain by species from the larger regional pool of species outside Britain, as a result of climate change, would increase bird diversity in what are now high diversity areas in the south, whether they are saturated or not under present conditions. There is now evidence that British bird species are already shifting the northern edges of their distributions further north (Thomas & Lennon 1999).

Status of the species diversity–energy hypothesis

Our results support the hypothesis. However, our results do indicate that the direct mechanism should be rejected. Although temperature was found to be the strongest determinant of diversity, the shifting pattern of associations was not as predicted by the direct mechanism; the predicted summer–winter switch did not occur. Nor was there evidence for an effect of body mass: there was no tendency for lighter species to show stronger associations with temperature. Taking the results as a whole, although there was good evidence for a temperature–diversity relationship supporting the diversity–energy hypothesis, there was none for the direct mechanism. This leaves an indirect mechanism: how this operates remains unknown.

Climate change and bird diversity

While we have necessarily used, for simplicity, a non-interactive model, species’ interactions may turn out to be important for reorganization of species’ distributions under climatic change. Interactions are important in determining insect spatial distributions in simple experimental systems (Davis et al. 1998). Although there is a large literature on the individualistic responses to past climate change (Graham et al. 1996), the extent to which the redistribution of individual species is influenced by interactions with other species is largely unknown.

Our results suggest that the effects of environmental change on bird diversity will vary in magnitude from place to place. Specifically, we predict that a climatic change involving warmer summers will increase diversity throughout Britain, and will have the stronger impact on winter diversity. The magnitude of this increase will vary geographically, even if there is a completely uniform temperature rise across Britain. The real pattern of temperature increase will undoubtedly be more complex, so the expected rise in diversity, according to our model, will consist of two components: the basic sensitivity of the bird fauna in a locality to a temperature increase (Fig. 6), and the magnitude of warming actually occurring. This is particularly true of the north and west of Britain, areas currently associated with low bird diversity. In these areas, we may see a substantial change in the composition of the local bird fauna.


We thank the thousands of volunteers whose efforts in recording the bird distribution data made this work possible. Andrew Davis, Elli Groner, Stephen Hartley, Bill Kunin, Bryan Shorrocks and Chris Thomas provided helpful comments on earlier drafts. We are grateful to John Lawton for his constructive criticism and encouragement. Peter Rothery provided a useful statistical critique. We acknowledge the financial support provided by NERC through its TIGER (Terrestrial Initiative in Global Environmental Research) programme, award number GST/02647. J.R.G. Turner acknowledges grants from the Royal Society and the Academic Development Fund of the University of Leeds for providing computing equipment and databases. J.J. Lennon is supported by a University of Leeds Fellowship.

Received 18 October 1998; revision received 29 November 1999