* Correspondence: Ken Thompson, NERC Unit of Comparative Plant Ecology, University of Sheffield, Sheffield S10 2TN, UK (fax 0114 2760159; e-mail Ken.Thompson@sheffield.ac.uk).
1 Using data from a survey of over 10 000 1-m2 quadrats in a 3000-km2 area, we examined the relationship between abundance and range for the vascular plant flora of central England.
2 At the level of the whole landscape, abundance was not related to local, regional or national range. Local, regional and national range were closely related to each other.
3 At the level of the whole landscape, range was significantly and positively related to both niche breadth (expressed as the range of habitats exploited) and to habitat availability, although niche breadth appeared to be more important. Abundance was not related to niche breadth or habitat availability. Since specialist species are mainly confined to uncommon habitats (especially wetlands), we conclude that the relationship between range and niche breadth is not an artefact of widespread species passively sampling more habitats.
4 At the level of individual habitat types, significant positive relationships between range and abundance were common. These relationships remained after controlling for the effects of phylogeny. For predominantly annual weed communities, the relationship was linear, but for perennial communities it was markedly ‘upper triangular’, i.e. all combinations of range and abundance were found except wide range/low abundance. The evidence suggests that this difference can be attributed to the greater mobility of annual weeds.
One of the more widely supported generalizations emerging from ecology at the present time is that locally abundant species tend to be widespread and locally rare species tend to be narrowly distributed (Gaston 1996; Gaston et al. 1997). Although this relationship has been reported from a variety of taxa, relatively few of the published examples concern plants (Brown 1984; Rapoport et al. 1986; Gotelli & Simberloff 1987; Söderström 1989; Collins & Glen 1990; Hanski et al. 1993; Rees 1995). Nor has the relationship been found on all the occasions that it has been sought for plants (Hanski et al. 1993). In this study we used extensive published (Grime et al. 1988; Hodgson et al. 1995) and unpublished survey data from central England to investigate this pattern further. In particular, we attempted to answer the following questions for a plant assemblage. (i) Is a positive relationship between range and abundance found at the whole landscape level, or only within more narrowly defined habitats? (ii) Is the relationship found in all habitats, and is the form of the relationship similar in all habitats? (iii) Do our results allow us to add anything useful to the debate about the causes underlying the relationship?
Data and definitions
Survey methods are described in detail in Hodgson (1986) and Grime et al. (1988). Briefly, identities and rooted frequencies of all higher plant species were recorded in over 10 000 1-m2 quadrats in a 3000-km2 area of central England. For each species in each quadrat, the number of 10 × 10 cm subdivisions in which the species was rooted was counted, and this number converted to percentage frequency. Mean rooted frequency of a species is the mean of occupied quadrats only. The use of rooted frequency ensures less seasonal variation than estimates of cover. Quadrats were deliberately placed to include all the major habitat types of the region, and a significant proportion were targeted on known localities of rare plants. Every quadrat was allocated to one of the 32 terminal categories (Grime et al. 1988) of a detailed hierarchical habitat classification system. The survey was designed specifically to record the herbaceous (and dwarf shrub) flora and did not adequately record the distribution or abundance of trees and large shrubs.
A recurrent concern of animal ecologists is that range–abundance correlations could be sampling artefacts, simply because species that occur at low densities will tend to be recorded from fewer localities at which they actually occur than species that occur at high densities (Gaston & Lawton 1990; Hanski et al. 1993; Gaston 1994). It is worth emphasizing at the outset that for our data this possibility can be discounted. The floras of Britain as a whole and the survey area in particular are well documented, and apparently narrowly distributed species have not simply been overlooked. Note that many quadrats were directly targeted on rare plants; no realistic programme of random sampling would have encountered more than a tiny minority of such species. It is possible that this could have artificially inflated the abundance of rare species, if locally abundant populations were sampled in preference to locally scarce ones. However, this certainly did not occur in the case of the rarest species, for which all known populations were sampled, and also seems unlikely to have been a problem for the less rare species.
Throughout this paper, abundance of a species means rooted frequency in percentage at the 1-m2 scale, measured in a 3000-km2 area of central England, as described above. We used three measures of range at the whole landscape scale. National range is the number of 10 × 10 km grid squares (hectads) occupied by a species in the British Isles. Regional range is the number of hectads occupied in north-central Britain. Local range is the number of 1-km2 grid squares occupied in the survey area. When we consider patterns at the habitat level, where range is much more a matter of occurrence in discrete but variably sized habitat patches, range is the number of quadrats occupied in the survey area. Before statistical analysis, we transformed both range and abundance, often but not always logarithmically, in an attempt (not always wholly successful) to homogenize variance.
Theoretically, positive interspecific relationships between abundance and range could arise from the shared ancestry of the species in the assemblage under consideration. This is most likely to occur if range and/or abundance are strongly associated with phylogeny, i.e. if much of the variation in these attributes is found at a high taxonomic level. Peat & Fitter (1994) have already shown that for the British range this is not the case for plants (Table 1); indeed British range is a relatively unusual example of a trait that varies almost entirely at the species level. We conducted a nested anova of abundance and found that just over half of the variation occurs at the species level. The remainder is spread rather evenly between subclasses, families and genera (Table 1). Gaston et al. (1997) also reported that whenever phylogenetically controlled analyses have been performed, significant relationships between abundance and range (in birds, mammals and macrolepidoptera) have not arisen from the non-independence of data points. Therefore we are probably justified in assuming that phylogenetic constraints make little contribution to the relationships described later in this paper. Nevertheless, we examined the possibility that these relationships are at least partly due to an association between abundance and phylogeny, by conducting phylogenetically independent analyses.
Table 1. Percentage variance at each taxonomic level for abundance and British range
We regressed abundance against range at the national, regional and local scales, which are themselves highly correlated (Table 2 and Fig. 1). At the national and local scales there was no evidence of any interspecific abundance–range size relationship. At the regional scale there was a weak negative relationship, but the amount of variation accounted for was trivial. At the level of the whole landscape there was therefore no evidence of any relationship between range and abundance.
Table 2. Non-parametric correlation coefficients between range in the UCPE survey area (local range), in north-central England (regional range) and in the British Isles (national range). n = 820. All significant at P < 0.001
Perhaps this finding is not too surprising. The landscape of the survey area is a patchwork, with the broad pattern of patches arising from variation in geology, topography and climate, but their detailed size and distribution is chiefly a consequence of human activity. We might therefore expect range to be largely determined by the available area of suitable patches. Wide-ranging species will occupy abundant habitats. Note that this idea relates to two distinct hypotheses that have been proposed as explanations for the interspecific abundance–range size relationship: the ‘niche breadth’ hypothesis (Brown 1984) and the ‘resource availability’ hypothesis (Hanski et al. 1993; Gaston 1994; Venier & Fahrig 1996). The former hypothesis assumes that widespread (and, less plausibly, locally abundant) species have an ability to use a broader spectrum of resources. The latter assumes that widespread (and locally abundant) species use widespread (and locally abundant) resources. As Gaston et al. (1997) observed, these two hypotheses are often conflated, are not necessarily independent, and are sometimes difficult to distinguish. Strictly speaking, we cannot test either hypothesis, since we do not have a significant abundance–range size relationship to explain, but nevertheless we can attempt to test some of their assumptions and predictions about range, abundance, niche breadth and resource availability.
One measure of (realized) niche breadth is the range of habitats utilized by a species. We have attempted to quantify this by reference to the hierarchical habitat classification of the survey quadrats. Seven broad habitat classes can be recognized: skeletal, arable, pasture, spoil, wasteland, woodland and wetland, each of which (except arable) is an amalgamation of a number of the terminal habitats described earlier. Habitat definitions can be found in Grime et al. (1988). For each species, relative frequency in each of these broad habitats can be assigned to one of five classes, where 5 is very common and characteristic of the particular habitat (percentage frequency >4 times that in the survey as a whole) and 1 is largely absent from the habitat (percentage frequency <0.25 times that in the survey as a whole). For each species, these seven scores can form the basis of a ‘specialism index’, which is the sum of the absolute differences of each score from the median value of 3. Thus a species that is equally frequent in all seven broad habitat types would have seven scores of 3 and a specialism index of zero. In fact, only one species (Poa trivialis) achieves this. At the opposite extreme, a species that is largely restricted to one habitat type will have one score of 5 and six scores of 1, giving a maximum specialism index of 14. Most species will lie somewhere between these extremes. We do not consider whether the success of generalists in a range of habitats arises from ecotypic differentiation or from phenotypic plasticity. Both may be involved (Snaydon & Bradshaw 1961; Sultan et al. in press). The specialism index was not calculated for species that occurred in fewer than five quadrats.
In Fig. 2 we plotted local range and abundance against specialism index. The relationship with range was highly significant and explains about 35% of the variation in the data (Fig. 2a), providing strong support for the hypothesis that widespread species use a wide variety of habitat types. In contrast, there was a weak but significant tendency for more specialized species to be locally more abundant (Fig. 2b).
In principle, testing the hypothesis that range is related to resource (i.e. habitat) availability is simple. In practice there are two serious difficulties. First, definition of the ‘habitat’ of a species is difficult, because species vary greatly in their habitat specificity. We therefore resorted to determining the most common terminal habitats, listed in Grime et al. (1988). The most common terminal habitat is that in which the species is most frequent, i.e. occurs in the greatest proportion of quadrats. Note that this habitat will be a good guide to the habitat preferences of specialist species, but an increasingly poor one for species of wide ecological amplitude. Secondly, determination of the extent of habitats within the survey area is difficult. The 10 000 survey quadrats themselves provide one estimate of habitat abundance, but of course the use of such data would generate an inevitable circular relationship. We therefore used data from the 25-class version of the Institute of Terrestrial Ecology (ITE) Land Cover Map of Great Britain (Fuller et al. 1994), although in fact only 19 habitat classes occurred in the survey area. Each terminal habitat was then linked, wherever possible, to one or more ITE land cover types, and thus to a corresponding percentage of the survey area. For example, the terminal habitats ‘arable’ and ‘unshaded mire’ were linked to the ITE classes ‘tilled land’ and ‘rough/ marsh grass’ plus ‘lowland bog’, respectively. It was not possible to link some terminal habitats, e.g. ‘hedges’ and ‘walls’, to any ITE cover type. For each terminal habitat that was the most common for at least 10 species, a mean local range and abundance was then calculated for those species, using data from the whole survey area. Mean range and abundance were then plotted against percentage of the survey area occupied by the habitat (Fig. 3). Note that this approach involves many approximations, both in the estimation of species’ habitats and in the area occupied by these habitats in the survey area, and is therefore very conservative.
There was a significant positive relationship between local range and habitat availability (Fig. 3a); widespread species occupy widespread habitats. Moreover, the true relationship may be stronger than that shown. Two habitats, road verges and acidic unenclosed pasture, stand out because although they are very abundant, they contain species with rather small ranges. The road verge habitat is linked to the ITE cover type ‘meadow/verge/semi-natural grass’, which was abundant in the survey area. However, most of the species that achieve their highest frequency in road verges are not generally common in other types of grassland, and therefore Fig. 3(a) overestimates the area available for these species. No simple explanation can be offered for the low mean range of species of acidic pasture.
In contrast, there was no relationship between abundance and habitat availability; locally abundant species do not tend to occur in widespread habitats. Indeed there was a non-significant tendency for the reverse to be true (Fig. 3b; see also Fig. 2b). In Fig. 3(c) we tested the obvious remaining question: Are plants of narrowly distributed habitats more specialized than plants of widespread habitats? The answer is a qualified ‘yes’. The overall relationship in Fig. 3(c) is not statistically significant, but two habitats stand out as being both rare and occupied largely by specialists. These are ‘lakes, canals, ponds and ditches’ and ‘unshaded mire’, the only two wetland habitats in Fig. 3(c). Thus while no pattern is evident in the terrestrial habitats, there is a clear tendency for plants of wetland habitats (which were rare in the survey area) not to occur outside wetlands and thus to have small ranges in the survey area.
If, as Fig. 3(c) suggests, niche breadth and habitat availability are not independent, which has the largest effect on range? We can test this by a step-wise multiple regression of both variables on range for the 401 species for which both are known. The specialism index enters the regression first and accounts for 26.5% of the variance in the data, while habitat availability accounts for only a further 1.5% of the variation. Thus, to the extent that the specialism index is a measure of niche breadth, this is a more important determinant of range than habitat availability. Neither is related to abundance.
Habitat level patterns
There was no evidence of any relationship between range and abundance at the landscape level. Does the position change at the level of individual habitats, and how finely do we have to focus in order to detect any pattern? We can begin to answer this question by regressing abundance on range for each of the seven broad habitat types described above. The results (not shown) do not differ appreciably from those at the whole landscape scale. The regressions are either not significant or (rarely) marginally significant and the proportions of variance explained are negligible (c. 1% in most cases). The lack of any relationship suggests that the broad habitat types are still too diverse, and that the processes determining range (i.e. habitat patch occupancy) are different from those determining abundance within patches, as described above for the landscape scale.
If we focus on the terminal habitats of the hierarchical habitat classification, a different picture emerges. Significant positive relationships are found in nearly every case, although the proportion of variance explained varies considerably. Figure 4 shows some typical examples. In plant communities that consist entirely or largely of perennials (Fig. 4a–e), there is a distinctly ‘upper triangular’ pattern, in which all combinations of range and abundance seem possible except high range/low abundance. In many cases the regression explains about 20% of the variance, but in some cases the proportion is much lower. In communities of annual plants (Fig. 4f–g) there is a more straightforward linear relationship. In fallow arable communities, which consist largely of annual weeds but have a perennial component, the pattern is intermediate, but closer to that in annual communities (Fig. 4h).
We also conducted separate phylogenetically independent analyses of the data for each community in Fig. 4, using a modified version of the CAIC (Comparative Analysis by Independent Contrasts) package for the Apple Macintosh computer (Purvis & Rambaut 1995). This new package runs under the Microsoft Windows v3.1 Graphical User Interface for IBM-compatible personal computers and performs the same calculations as the original program. Standardized linear contrasts were calculated using the Crunch procedure designed for continuous variables. Further details of the program and of the phylogeny employed can be found in Hodkinson et al. (in press). We then regressed the standardized linear contrasts of abundance against the contrasts for range, forcing the regression through the origin (Purvis & Rambaut 1995). The results of these analyses are summarized in Table 3. Strictly speaking, one cannot compare r2 values from regressions forced through the origin with those from unconstrained regressions. Nevertheless, the correspondence between the regressions in Fig. 4 and Table 3 is striking, suggesting that the relationship between range and abundance at the habitat level owes little or nothing to phylogeny.
Table 3. Regression analysis of standardized linear phylogenetically independent contrasts between range and abundance in eight contrasted plant communities. In each case regression was forced through the origin. n = number of contrasts
Unenclosed limestone pasture
Woodland on limestone
Unenclosed pasture on acidic substrata
Riverbanks on calcareous substrata
Previous reports of a positive interspecific abundance–range size relationship in plants have tended to concentrate on quite narrowly defined habitat types; for example prairies (Gotelli & Simberloff 1987; Collins & Glen 1990), bryophytes on logs (Söderström 1989), sand dune annuals (Rees 1995) and coastal meadows (Hanski et al. 1993). Previously, only Rapoport et al. (1986) had reported a positive relationship from a diverse landscape (Berkshire), but their study employed an unknown, ‘random’ sample of only 80 species and measured abundance at a very coarse scale. The results presented here are therefore the first well-documented study of the entire flora of a diverse landscape at all scales from individual habitats up to the whole landscape. They allow a new perspective on some of the hypotheses proposed to account for positive abundance–range size relationships.
As Gaston et al. (1997) observed, neither the niche breadth nor the resource availability hypothesis has much difficulty in explaining why species that utilize widespread resources are themselves widespread. They find it much harder to explain why such species should also be locally abundant. Essentially both hypotheses require that widespread resources should also be locally abundant. The present study shows that, at the landscape scale, this is clearly not the case. Resources (habitats) suitable for, say, wetland plants may well be locally abundant where they occur, yet scarce in the landscape as a whole. It is therefore not at all surprising that no range–abundance relationship can be detected in a diverse, patchy landscape. Our data indicate that the distribution of plants can be explained, at least partly, by the distribution of habitats. Abundance cannot be explained in such terms, and we have deliberately avoided the difficult topic of exactly what does determine abundance in plants. Why some plants routinely attain high abundance, but others rarely if ever do so, is a complex subject with its own large literature (for example, Grime 1987).
Breadth of resource usage appears to explain quite a lot of variation in range (Fig. 2a). However, as Burgman (1989) has elegantly demonstrated, such apparently convincing patterns can easily arise from sampling bias. Even if species sample the landscape at random, widely distributed species will occur in a wider range of habitats (and thus appear to be more generalist) than narrowly distributed species. Fortunately, however, sampling bias makes some quite specific predictions about the observed pattern of specialism, and these predictions are easily tested. If the pattern in Fig. 2(a) is a sampling artefact, then narrowly distributed species will be more frequent in common habitats. Thus most apparent specialists will be plants of arable fields or various types of grassland, which overwhelmingly dominate the survey area. In contrast, if the pattern has a real biological explanation, we would not expect specialist species to be confined to common habitats. It is less clear what we would expect; perhaps the most realistic expectation is that specialists would be most frequent in the most distinctive habitats, i.e. those habitats that differ most from the majority of the survey area. Figure 2(b) is entirely consistent with this latter prediction. In a wide variety of terrestrial habitats, there is no relation between habitat area and the mean specialism index of the species present; specialists are not confined to common habitats. Habitats that are rare but distinctive, i.e. mires and (especially) water, are characterized by a highly specialized flora. This interpretation of the data is confirmed by an analysis of unique species, i.e. those that occur only in one narrowly defined habitat type. For the habitats analysed in Fig. 3, this new analysis confirms that wetlands, despite their limited extent, contain many unique species, while the number of unique species in terrestrial habitats is unrelated to habitat area (Fig. 5). We therefore think it is unlikely that the pattern in Fig. 2(a) is an artefact.
At the level of individual habitats, a positive abundance–range size relationship seems universal in this study. Can this pattern be accounted for by a correspondence between the large scale and local abundance of habitats? Here we need to digress briefly into what is meant by ‘habitat availability’. Plants, say in limestone grassland, have unique tolerances along the various axes that make up the limestone grassland habitat. These axes include (inter alia) grazing intensity, altitude, aspect, soil moisture and depth, pH and other mineral nutrients and incidence of fire. Many of these variables will vary both within and between habitat patches, but some, particularly past and present management practices, will be relatively uniform within patches. Some species, either because their particular requirements are widespread or because they have wide tolerances, will find their requirements met in more limestone grassland patches than other species. However, since most environmental variables vary within patches as well as between them, plants that find their requirements met in many patches are very likely to find them met over a large part of individual patches. This satisfactorily explains why, within a sufficiently narrowly defined habitat type, widespread plants are also locally abundant (Fig. 4). Pursuing the example of limestone grassland, it also explains why some scarce species (at least in perennial plant communities) are locally abundant. Some species, either because their particular requirements are scarce or because they have narrow tolerances, will find their requirements met in relatively few limestone grassland patches. The factors that make these few patches suitable, which may involve any of the variables listed above, may also be scarce within patches. Alternatively, these factors may be scarce at the landscape scale but common within patches. This latter situation is particularly likely to apply to site factors that arise from an unusual site history; there is a good deal of evidence that the occurrence of uncommon plants is often related to peculiarities of site history rather than any obvious modern features (Pigott & Walters 1954; Rackham 1980). Thus narrowly distributed species may be either locally abundant or not. The only forbidden combination of range and abundance is wide range and low abundance, which would require habitat features that are widespread in the landscape but always rare within individual patches. It is hard to imagine what such features would be, and it is therefore not surprising that we do not observe widespread plants with low abundance. Rabinowitz (1981) described plants of this sort as exhibiting ‘the most curious form of rarity’, and suggested that such plants were frequent in North America, for example sparse prairie grasses. It is not clear whether Rabinowitz's opinion, which seems to be based on anecdotal evidence, would be supported by objective evidence. Certainly Gotelli & Simberloff's (1987) data, from tallgrass prairie in Kansas, seem to show a complete absence of widespread but locally abundant species.
An alternative explanation for the existence of numerous locally abundant but narrowly distributed perennials is that the extant populations of these species are the remnants of formerly widespread distributions. We do not consider this possibility further here, but the fact that many British plants have suffered major recent reductions in range suggests that this may be a contributory factor in at least some species (Thompson & Hodgson 1996). It is also worth noting that the whole subject of abundance in ‘patches’ is very scale-dependent; see, for example, Pearman's (1997) discussion of the dependence of rarity and rates of decline on whether one measures distribution at the level of hectads, tetrads, 1-km2 squares or individual populations.
The existence of many scarce but locally abundant species (Fig. 4) also makes the various metapopulation hypotheses unlikely, at least in communities of perennial plants. All variants of these hypotheses (Hanski et al. 1993; Gaston et al. 1997) require that locally more abundant species will always occupy more patches at equilibrium, although clearly this will depend to a large extent on relative dispersal capacities. We explicitly consider the role of dispersal in a subsequent publication, but at present metapopulation hypotheses do not appear to be consistent with most of the data from this study. Metapopulation processes do, however, seem more important for annual plant communities (see below).
The annual plant communities of arable fields and related habitats do not conform to the above pattern, in that they do not contain narrowly distributed species with high abundance (Fig. 4). Two explanations for this difference seem likely. First, it may be that arable fields, as a result of massive human intervention, are much less heterogeneous (both within and between patches) than semi-natural habitats. Much effort has been expended, after all, in attempting to make them optimal for the growth of one or other of a handful of fast-growing cultivars. Arable weeds are therefore likely to be either well adapted to the arable habitat, and therefore both widespread and abundant, or less well adapted, and therefore neither widespread nor abundant. A species that is successful in one wheat field will probably find its requirements met by most wheat fields. A second explanation depends on the life histories of arable weeds. Successful annual weeds must possess a number of traits, including a phenology that fits with the timing of the prevailing crop sowing and harvesting regime and, increasingly in the modern landscape, herbicide resistance. However, an annual weed that is abundant in one or a few arable fields must also by definition possess a high seed output. Given the substantial transfer of seed between fields in soil and on vehicles (McCanny & Cavers 1988; Bakker 1989; Milberg 1991; Prach et al. 1995; Hodkinson & Thompson 1997), it is inevitable that such species will spread rapidly to other fields and therefore become widespread. In other words, the life-history traits that confer success in a particular patch would inevitably lead to spread into other patches. In this sense, arable weeds may be behaving more like animals, in tending to disperse away from high-density patches.
These two hypotheses lead to readily testable predictions. If the absence of narrowly distributed species with high abundance is a consequence of habitat uniformity, the same pattern should be found in other uniform habitats, even those occupied by perennials. If, on the other hand, this absence is due to the high seed output and mobility of annual weeds, uniform habitats occupied by perennials should show the ‘upper triangular’ pattern. We tested these ideas by analysing data from lowland, improved ryegrass pastures. These communities, as a consequence of drainage, reseeding, fertilizer and herbicides, have been rendered relatively uniform, but contain an almost exclusively perennial vegetation. The results (Fig. 6) conform to the ‘upper triangular’ pattern, suggesting that the mobility of annual weeds is mainly responsible for the observed patterns. It therefore seems possible that metapopulation hypotheses may be more important in communities of short-lived, mobile species than in communities of perennial plants.
We conclude that the prevalence in the literature of positive abundance–range size relationships for plants is a consequence of a concentration of previous work in relatively uniform plant communities. When a diverse, patchy landscape is investigated, no such relationship is evident at the landscape scale. At this scale, range is strongly linked to habitat availability, which itself appears to be largely a consequence of range of habitat tolerance. This finding does not, of course, preclude the possibility (indeed probability) that in other landscapes highly specialized species might be widespread, if their preferred habitat happened to be widespread. At the scale of the individual habitat type, positive abundance–range size relationships are common. We provisionally explain the ‘upper triangular’ shape of most of these relationships as a consequence of the usual, but not universal, ‘self-similarity’ of key habitat variables, i.e. such variables are common or rare at all spatial scales. The absence of narrowly distributed species with high abundance in arable weed communities is best explained by the greater mobility of such species.
This research was supported by the Natural Environment Research Council, through its Large Scale Processes in Ecology and Hydrology programme, grant number GST/02/1211. We thank Sue Wallis of the Institute of Terrestrial Ecology at Monks Wood for the provision of land cover map data, Henry Ford for invaluable assistance with the Ecological Flora Database, and two referees for comments on an earlier version. K. J. Gaston is a Royal Society University Research Fellow.
Received 21 April 1997revision accepted 3 November 1997