SEARCH

SEARCH BY CITATION

Keywords:

  • area of occupancy;
  • change;
  • distribution pattern;
  • dynamism;
  • fractal

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information
  • 1
    The distribution patterns of 391 rare and scarce British plants (species recorded in 100 or fewer 10 × 10 km squares) were characterized by their distributional area (area of occupancy at 1-km scale: AOO1) and levels of aggregation (as reflected in fractal dimensions measured across two scales: D1−10 and D10−100).
  • 2
    Eighteen plant traits were tested for relationships to AOO, and to fractal dimension while controlling for AOO. These included both directly heritable traits (e.g. life-form) and emergent properties that are, at most, indirectly heritable (e.g. typical local density). The latter set included an index of net distributional change and an index of range dynamism.
  • 3
    Only two traits, habitat preference and local abundance, were significantly related to AOO1, but about half were associated with fractal dimension.
  • 4
    Relatively aggregated fine-scale distributions (high D1−10) were related to high local abundance, lack of specialized, long-distance dispersal mechanisms, habitat preference and an increasing range size with relatively few local extinctions (i.e. a positive index of change with low dynamism).
  • 5
    Relatively aggregated coarse-scale distributions (high D10−100) were related to the use of insect pollinators, obligate outcrossing, habitat preference and relatively stable ranges (low dynamism).
  • 6
    Multivariate analyses of subsets of conceptually related variables showed that few variables interacted to affect distributional variables.
  • 7
    A highly significant negative relationship between dynamism and fractal dimension appears to be driven primarily by high rates of local extinction, leading to relatively scattered, diffuse range structures. Furthermore, it suggests that recent population trends may be inferred from snapshots of contemporary distribution patterns.
  • 8
    The role and interpretation of phylogenetically informed analyses in studies such as this are debatable. However, we found similar relationships in both phylogenetically informed and conventional analyses for all variables except pollination vector (a strongly conserved trait).
  • 9
    The spatial pattern of plant species distributions is associated with a range of ecological traits, particularly those describing past changes in distribution. The analysis of distribution patterns therefore has the potential to inform future conservation effort.

Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

Understanding the abundance and spatial distribution of species is one of the fundamental goals of ecology (Andrewartha & Birch 1954; MacArthur 1972; Brown 1995). To explain fully the population patterns of even a single species is a substantial task, requiring an intimate knowledge of the species’ natural history, its abiotic limitations and its full range of biotic interactions. An alternative (and more often attempted) approach is to examine statistically the distributions of a large number of species, looking for traits that are correlated with patterns of abundance or distribution (Gaston & Blackburn 2000; Cadotte et al. 2006). Several studies have compared the traits of sets of rare and common species (reviewed by Kunin & Gaston 1993; Gaston & Kunin 1997), and some generalizations have emerged, with, for example, higher levels of asexuality, higher levels of homozygosity and lower reproductive investment being frequently reported in rare species (e.g. references from Kunin & Gaston 1997; Hegde & Ellstrand 1999; Cadotte & Lovett-Doust 2002; Kessler 2002; Lloyd et al. 2002). Nonetheless, it has been striking how few consistent differences are found to explain differences in abundance. In part, this may be because specific measures of abundance may miss much of the richness of species distributional patterns. In this paper we therefore consider the relationships between plant traits and the patterns of their distributions.

There are many indices used to describe spatial distribution patterns (reviewed by Dale 1999). Hartley & Kunin (2003) suggested that the combination of two indices, area of occupancy (AOO), measured at a specified scale (as it is inherently scale-dependent), and the box-counting fractal dimension (DB), measured over a specified range of scales, could describe the major aspects of range structure over a range of scales. The first of these is familiar, as most studies of species commonness use counts of occupied grid cells, at some convenient scale. The addition of a fractal dimension incorporates information about the dispersion of occupied cells (Halley et al. 2004). The box-counting fractal dimension describes the way in which AOO varies as a function of the scale of measurement, and can be calculated from the slope of scale–area curves (Kunin 1998; also termed ‘range-area relationships’; Ostling et al. 2003) of species distributions. These are easily calculated, for coarse scales, from typical atlas ‘dot maps’ (e.g. Preston et al. 2002). In analyses of a finite area, fractal dimension is related to AOO; a single species occurrence is a single point at all resolutions and has a fractal dimension of zero, while a solid area (at all resolutions) has a fractal dimension of two (Halley et al. 2004). Nonetheless, for intermediate levels of AOO, there is a wide range of possible DB values. Among these, a relatively high fractal dimension indicates an aggregated distribution, whereas a low fractal dimension indicates a scattered distribution.

Traits may affect distribution patterns in two ways: traits may cause a particular range structure or they may permit it (cf. Kunin & Gaston 1997). For example, in clonal plants, vegetative reproduction may cause fine-scale aggregation in the species distribution. Longevity, by contrast, does not cause a scattered distribution, but may permit a more scattered distribution than would be possible in short-lived species, because of the persistence of isolated individuals outside the range within which recruitment is currently possible. Whatever the cause, the traits associated with such spatial patterning are of substantial interest for conservation biologists and managers.

Detailed analyses of species distributional properties require large and detailed databases. Such databases have only become possible due to the co-ordinated efforts of a large number of professional and volunteer biologists collecting standardized distributional data for large taxonomic assemblages. We selected to use distribution information on the rare and scarce plants in Britain (the 391 taxa found in fewer than 100 10 × 10 km squares; Stewart et al. 1994; Wigginton 1999) because their fine-scale distribution is well known. For the current analysis we used high-quality information on the recorded locations of these taxa, most of which are recorded to 100-m resolution, although, to be conservative, we considered them to be precise only to the nearest 1 km (Biological Records Centre, CEH Monks Wood, UK). In restricting ourselves to rare and scarce taxa we are of course prevented from making comparisons between rare and common taxa. However, many such analyses have already been performed by others (see References above) and the high quality of the distributional information available for these rare and scarce plant species makes possible detailed analyses of distributional patterns and their biological correlates, which are to our knowledge unparalleled elsewhere.

Multi-species comparisons of this type run the risk of becoming data mining exercises, with post hoc rationalizations presented as if they were a priori predictions. To guard ourselves from this, we collected data on a range of variables considered to be of potential importance in explaining species distributions and abundance, and in each case formulated predictions based on the results of theoretical models and previous comparative studies (mainly studies of rare vs. common taxa, so our study provides a different context within which to test the generality of these results). The following specific hypotheses were tested in this study:

  • 1
    Life-form. Raunkiaer life-form classes categorize plants according to their life history and growth form, which is strongly associated with longevity (Clapham et al. 1952). Increased longevity is expected to allow plants to persist in larger and more scattered ranges (Kelly & Woodward 1996; Kelly 1996); therefore, perennials (phanerophytes and chaemophytes) are expected to have larger, more scattered ranges. Annuals (therophytes) are expected to display opposite patterns.
  • 2
    Woodiness. Woodiness is also an indirect measure of longevity, and thus woody plants are expected to have relatively large and more scattered ranges compared with non-woody plants (e.g. Harper 1979). The distribution of woody plants is also affected by physiological constraints, such as water stress, but these probably have little effect in Britain (Kelly 1996).
  • 3
    Local abundance. A positive relationship between regional distribution and local abundance is a commonly observed pattern (e.g. Blackburn et al. 1997; He & Gaston 2000). Hence, we expect locally abundant (dominant) taxa to have larger range sizes. If spatial patterning is correlated across scales, as appears to be the case (Kunin 1998; He & Gaston 2000; Kunin et al. 2000; but see Hartley et al. 2004), we would also expect locally abundant taxa to have relatively aggregated distributions (Hartley 1998; Holt et al. 2002).
  • 4
    Propagation method. Species that only reproduce vegetatively in the British Isles will commonly form cohesive, clonal patches at local scales, which we expect to be reflected in relatively aggregated ranges at larger scales (as with the hypotheses relating to local abundance). Species that propagate by seed have greater potential for widespread dispersal and persistence in a seed bank, which may permit them to have relatively scattered distributions. Those propagating via both methods might be expected to show features of both. Clonal plants (reproducing entirely vegetatively) are expected to have a smaller AOO than unitary plants, because they are deemed to be less ecologically flexible (Kelly & Woodward 1996).
  • 5
    Fertilization mode. It is expected that obligate outcrossing species have relatively aggregated distributions, because individuals must be in close proximity to unrelated conspecifics in order to reproduce successfully. Species that normally self-fertilize can form and persist in more scattered ranges, as an isolated individual can potentially establish a population after a long-distance dispersal event (Baker 1955; Quinn et al. 1994; Barrett et al. 1996). The presence of such outposts would increase overall AOO (particularly when measured at a coarse resolution) but would decrease DB.
  • 6
    Pollen vector. The lack of a pollen vector implies self-fertilization of seeds, which should permit species to persist in ranges with low fractal dimensions. In taxa in which obligate cross-pollination occurs, wind pollination requires plants to be relatively aggregated (Allison 1990). High local densities are also important for insect-mediated pollination (e.g. Platt et al. 1974; Klinkhamer et al. 1989; Kunin 1992, 1993, 1997), but because insects are potentially specific and efficient transfer mechanisms for pollen, we expect insect-pollinated species to persist with more scattered distributions. We had no specific prediction for AOO, although Harper (1979) found wind-pollinated plants to be under-represented in lists of rarities.
  • 7
    Seed mass and production. The size and number of seeds produced are subject to a trade-off in resource allocation in plants (Thompson et al. 2002) and have complex interactions with the mode of dispersal (Quinn et al. 1994; Edwards & Westoby 1996). For example, tiny seeds in the Orchidaceae can be produced in vast numbers and are dispersed by the wind over great distances. Larger seeds are more costly to produce and so are produced in smaller numbers, but very large seeds are often vertebrate-dispersed, and so may be dispersed over large distances, too. Long-distance dispersal (whether from tiny or large animal-dispersed seeds) should permit relatively scattered distributions, but we expect that small-seeded species are better dispersers overall and therefore have relatively larger, more scattered distributions.
  • 8
    Diaspore dispersal. Good dispersers will be able to disperse over much greater distances and develop distributions that are less aggregated (ignoring the possible effect of seed persistence in soil banks, i.e. ‘dispersal in time’; Thompson & Hodgson 1996). What constitutes a good disperser is a matter of debate (Quinn et al. 1994; Edwards & Westoby 1996; Kelly & Woodward 1996; Thompson & Hodgson 1996; Thompson et al. 2002), but we expect that plants with no specialized dispersal mechanisms have relatively aggregated and small distributions. We also expect that plants with ‘short-distance’ dispersal mechanisms (where the diaspores are dropped, dispersed explosively or dispersed by ants) have relatively small, aggregated distributions compared with those dispersed by wind, water or vertebrates.
  • 9
    Change in distribution. There are two plausible scenarios for the way in which changes in distribution could affect range structure. One is that species that are increasing in distribution do so in a cohesive (i.e. aggregated) way, whereas declining species leave scattered, relict populations. Alternatively, species expand their ranges primarily by a series of long-range jump dispersal events, and then the bulk of the population fills in behind the leading front by a slower, more diffuse process (Suarez et al. 2001), so that species with expanding distributions might be expected to display relatively scattered distribution patterns. Theoretical simulation studies have shown that either result is possible, depending upon the nature of the dispersal function and on the variation in species fitness across the landscape (J.J. Lennon, unpublished data). Nonetheless, a pattern of more diffuse distributions for declining species is predicted when realistic levels of habitat variation are modelled, and is supported by empirical studies of butterfly distributions (Wilson et al. 2004). An index of range expansion/contraction (‘change’; Telfer et al. 2002) was used to test this hypothesis. We also expected a positive association between distributional change and AOO.
  • 10
    Dynamism. Species distributions are not constant over time, owing to local extinctions and colonizations (e.g. Preston et al. 2002). This not only results in net changes in distributional extent (as discussed above), but it can also result in dynamic turnover among the remaining populations. We used a method for calculating distributional ‘dynamism’, or spatio-temporal turnover, independent of net change in species range size (Preston et al. 2003). As with the change index, two scenarios are plausible. Species that persist in relatively scattered ranges could be vulnerable to chance extinction events and so have highly dynamic distributions. Alternatively, species with highly dynamic ranges could show regular colonization events, which would generally be close to other occupied areas, and so have relatively clumped distributions. Whichever scenario is shown by the data, it is probable that the indices of dynamism and change are jointly influenced by the same underlying demographic processes.
  • 11
    Habitat. Habitat is predicted to exert a strong influence on plant distributions because the underlying factors affecting habitat (geology, rainfall, history of land use, etc.) show spatial structure (Klinkenberg & Goodchild 1992; McAlpine & Wotton 1993), which may constrain species distributions. Habitat has previously been shown to explain more of the variation in aggregation of rare and scarce plants than any other trait (Peat & Fitter 1994; Quinn et al. 1994). The prediction is that species distributions should reflect the AOO and fractal dimension of their habitat, but it is difficult to obtain an independent measure of this for rare and scarce plants. Although their requirements are sometimes well known, it is difficult to assess the national availability of suitable habitat for each species in the data set. For example, not every area of woodland identified by landcover surveys will be suitable for rare ‘woodland’ plants.

Some studies of the ecological correlates of plant distributions have assumed that each plant taxon is an independent data point (a so-called cross-species analysis). Other studies have included information about the relatedness of taxa (e.g. Harvey & Pagel 1991; Wright et al. 2000) to infer patterns of trait change over evolutionary time, allowing phylogenetically informed analyses. There has been much debate concerning the applicability of such approaches (e.g. Rees 1995; Westoby et al. 1995), so we compared results from both cross-species and phylogenetically informed approaches. We also specifically tested the dependence of the distributional characteristics on phylogeny (Freckleton et al. 2002). We conducted analyses of each plant trait individually, but also undertook multivariate analyses of subsets of conceptually related variables in order to assess the importance of interactions between the variables in affecting plant distribution patterns.

Methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

species data

The data set consisted of the rare and scarce plants in Britain (England, Scotland and Wales), i.e. those taxa recorded in 100 or fewer 10 × 10 km squares (hereafter referred to as 10-km squares) by Stewart et al. (1994) and Wigginton (1999). Recognition of taxonomic distinctions such as subspecies and varieties followed those in Stewart et al. (1994) and Wigginton (1999).

Data on the location of plants, precise to the nearest 1 km or finer and dated between 1987 and 1999 (Preston et al. 2002), were obtained from the Biological Records Centre, CEH Monks Wood. Only records that were ‘native’ or ‘assumed native’ were included in the data set. The study area was restricted to mainland Britain plus a few nearshore islands, so that species distribution patterns would not be influenced by the intermingled distribution of land and sea in an island archipelago. Specifically, records from the Outer Hebrides, Orkney, Shetland, the Isle of Man and the Isles of Scilly were excluded, but records from the Inner Hebrides, Anglesey and Isle of Wight were retained. Finally, very rare taxa, defined as those occupying fewer than five 1 × 1 km squares (hereafter referred to as 1-km squares) were excluded. The resulting data set consisted of 391 taxa, recorded from between 5 and 239 1-km squares and between 1 and 102 10-km squares. (Chamaemelum nobile was recorded in 102 10-km squares due to an increase in its recorded distribution since the publication of Stewart et al. 1994).

Life-history attributes and the local abundance of the species were obtained from the Ecological Flora database of the British Isles (Fitter & Peat 1994). Additional life-history data (< 10% of the total) were obtained from Biological Flora of the British Isles in Journal of Ecology and other literature sources (full details are available from the corresponding author). Habitat classifications were taken from Stewart et al. (1994) and Wigginton (1999). The change index is the standardized residual of the weighted linear regression of counts of grid cells between an earlier survey period (1930–69) and a later period (1987–99) (Telfer et al. 2002). It reflects the relative change in the number of 10-km cells occupied, taking account of recorder effort, and was obtained from Preston et al. (2002). In our data set, the change index varied around zero (no change) from −2.40 to +2.81. The index of distributional dynamism was calculated from the number of new 10-km squares colonized (gains) between the two survey periods as a proportion of the number finally present (i.e. in 1987–99), and the number of 10-km squares lost as a proportion to the number initially present (i.e. in 1930–69) (Fig. 1; Preston et al. 2003):

image

Figure 1. Isoclines of dynamism of rare and scarce British plants (calculated from the rate of extinction and colonization, between the 1930–69 and 1987–99 surveys, as explained in the text). Dotted lines show the 0.2, 0.4, 0.6 and 0.8 isoclines. The proportional rate of colonization is the number of apparent colonizations per 10-km square occupied in 1987–99. The proportional rate of extinction is the number of apparent extinctions per 10-km square occupied in 1930–69. The change index is the perpendicular distance from the line of no change (solid), once corrected for recorder effort (Telfer et al. 2002).

Download figure to PowerPoint

  • image(eqn 1)

where g and l are the proportional gain and loss, respectively. The denominator scales the index between zero (no change in the identity of occupied grid cells) and one (constant AOO, yet the identity of the occupied cells changes completely). In our data set, the dynamism index varied from 0.00 to 0.89. Both change and dynamism indices were calculated using total recorded distributions, and thus included non-native occurrences. The indices were designed to be orthogonal to each other, and in our data set they were only weakly correlated (r = 0.098, P = 0.093).

Both indices are constrained by the availability of data from the two surveys. There was potentially an under-representation of the true value of the change index for taxa that showed change within a survey period (which becomes increasingly likely as the length of the survey period increases). The dynamism index was potentially under-represented for taxa with rapidly dynamic distributions (so that their recorded presence over a multiple-year survey period is a gross overestimate of their annual distribution). The indices are also subject to spatio-temporal differences in recorder effort, and although temporal differences are accounted for at a national level, there may still be regional changes in recorder effort affecting both indices. However, the lack of a significant association between the indices and biological traits, such as Raunkiaer life-form or propagation mode (results not shown), suggest that biases are not systematic and that relationships that we identify between change and dynamism and distribution patterns are real.

The attributes used in the analysis are shown in Table 1. Where possible (e.g. with woodiness) categorical variables were treated as ordinal variables. Pairwise associations between the variables were examined using non-parametric tests. Spearman's rank correlation coefficient was used to test for associations between two continuous variables, Kruskal–Wallis’H (with 10 000 Monte Carlo simulations) for associations between a continuous and a categorical variable, Kendall's τb (exact test) for an ordinal and an ordinal or dichotomous variable, and Fisher's exact test for two dichotomous variables. Sequential Bonferroni correction (Quinn & Keough 2002) was applied to the results of these tests. Variables that resulted from the reclassification of the same data (e.g. dispersal mode and long/short dispersal) were not tested against each other. Pairwise associations involving life-form or habitat were not tested because the large number of classes resulted in a low number of species in each two-way cross-classification.

Table 1.  Attributes used in the analysis. Dependent variables are indicated by ‘D’, and independent variables are numbered (those sharing the same number are derived from one another). Data types are indicated as: C, continuous; N, nominal (unordered categorical); O, ordinal (ordered categorical); B, binary. The group to which each variable was assigned for multivariate analyses is also indicated. Values for ordinal and binary variables are given in order to interpret the slopes of the regressions presented in this paper
 AttributeData typeMultivariate groupCategories n
DAOO1C Log10-transformed area of occupancy at 1 km391
DAOO10C Log10-transformed area of occupancy at 10 km391
D D 1−10 C Fine-scale fractal dimension (1–10 km)391
D D 10−100 C Coarse-scale fractal dimension (10–100 km)391
DResidD1−10C Residual of the regression between D1−10 and AOO1391
DResidD10−100C Residual of the regression between D10−100 and AOO10391
1Raunkiaer life-formN1phanerophyte; chamaeophyte; hemicryptophyte; geophyte; helophyte; hydrophyte; therophyte391
2WoodinessO11 = not woody; 2 = woody at base; 3 = woody391
3Local abundanceO2, 3, 51 = scattered; 2 = frequent; 3 = dominant219
4Propagation modeO21 = vegetative; 2 = seed and vegetative; 3 = seed266
5aFertilization modeO31 = obligate or normally outcrossing; 2 = both cross- and self-fertilization; 3 = self-fertilization or apomictic208
5bObligate outcrossingB30 = self-fertilization, normally outcrossing or both; 1 = obligate outcrossing208
6aType of pollen vectorN3none; insect; water; wind342
6bPresence of pollen vectorB30 = none; 1 = insect, water or wind342
6cInsect pollinatedB30 = wind or water; 1 = insect304
7Seed massC4Log-transformed seed mass (measured in mg)126
8Seed production (per flower)O41 = 1 seed; 2 = 1–10 seeds; 3 = 10–100 seeds, 4 = 100–1000 seeds; 5 = > 1000 seeds117
9Seed production (per plant)O41 = 1–10 seeds; 2 = 10–100 seeds; 3 = 100–1000 seeds; 4 = 1000–10000 seeds; 5 = > 10000 seeds 69
10aDiaspore dispersalN4unspecialized; explosive; ants; water; wind; vertebrates (by ingesting, epizoochory or caching)148
10bDispersal (specialized vs. unspecialized)B40 = unspecialized; 1 = all other types of dispersal148
10cDispersal (long vs. short)B40 = unspecialized, explosive or ants; 1 = wind, water or vertebrates148
11Change indexC5(see text for further details)292
12Dynamism indexC5(see text for further details)292
13HabitatNNoneA = arable; CG = basic grassland and rock outcrops; C = coastal, including sand dunes, salt marshes and brackish water; D = damp mud and water margins; F = calcareous flushes and fens; FW = fresh water; H = heaths, moors and bogs; M = montane; NG = neutral and acid grassland; W = woods, hedges and scrub.365

calculation of distribution indices

To characterize the distribution patterns of each species, we used three measurements: the log10-transformed area occupied at 1-km2 resolution (AOO1), the box-counting fractal dimension of the distribution between linear scales of 1 and 10 km (D1−10), and the box-counting fractal dimension between linear scales of 10 and 100 km (D10−100). We calculated the fractal dimension from the slope of a linear regression of log number of cells occupied against log area occupied, at resolutions of 1 and 10 km (D1−10) or 10 and 100 km (D10−100) (Peitgen et al. 1992; Hartley et al. 2004). The three measures can be visualized as the intercept of a scale–area curve (AOO1) and its slope over two successive ranges of scales (D1−10 and D10−100) (Kunin 1998; Hartley & Kunin 2003).

In natural data sets the box-counting fractal dimension is strongly affected by the number of squares occupied at the finest scale (Halley et al. 2004). Therefore, before analysing the effect of plant traits, the effect of the number of occupied squares on fractal dimension was tested. When tested separately, both AOO1 and D10−100 were significant predictors of D1−10 (AOO1: F1,389 = 16.87, P < 0.001; D10−100: F1,389 = 7.52, P = 0.006). However, when the two were included together only AOO1 had a significant effect (for the partial correlation: F1,388 = 9.20, P = 0.003; compared with D10−100: F1,388 = 0.05, P = 0.828). Similarly, AOO10 was a significant predictor of D10−100 (F1,389 = 172.09, P < 0.001). Therefore, in all the subsequent analyses AOO1 was included as a covariate when testing effects on D1−10, and AOO10 was included when testing effects on D10−100. The standardized residuals of the regressions between DB and AOO (ResidDB; Table 1) were retained for analysis where covariates could not be included and for the graphic presentation of results.

phylogenetic dependence

When considering the relationship between plant traits and distribution patterns it is important to consider the potential role of phylogeny. We constructed a phylogenetic ‘supertree’ for the taxa in the data set, based on the family-level phylogeny of Stevens (2002), developed from work of the Angiosperm Phylogeny Group (1998). Published phylogenies were used to resolve the tree below the family level, and where these were not available for the species of interest, taxonomic groups were assumed to be monophyletic and used as a surrogate for phylogeny (Garland et al. 1992). The full tree and further details of its construction are given in supplementary Appendix S1.

The degree of phylogenetic dependence in the dependent variables indicates whether phylogeny should be considered in the analysis (Freckleton et al. 2002). The variable λ indicates the degree of phylogenetic dependence, varying from zero (phylogenetic independence) to one (strong phylogenetic dependence). The value of λ was estimated for distributional variables (AOO1, AOO10, ResidD1−10 and ResidD10−100) using the program continuous (Pagel 1999), as described by Freckleton et al. (2002), with the phylogeny in Appendix S1 and path lengths set equal to one. Log-likelihood tests were used to test the hypotheses that λ = 0 and λ = 1. All four variables were significantly different from one, and three were significantly different from zero (Table 2), indicating some phylogenetic dependence in the variables. Fine-scale distributional variables (AOO1 and ResidD1−10) showed a greater degree of phylogenetic dependence than coarse-scale variables (AOO10 and ResidD10−100). Further examination showed that there was a positive relationship between the power of the log-likelihood tests and the size of the tree (Freckleton et al. 2002), and extrapolation implies that significant differences are almost inevitable with a tree as large as ours.

Table 2.  Phylogenetic dependence of range size and distribution patterns. The value of λ indicates the degree of phylogenetic dependence. The log likelihood (ln lik) of the estimated value is shown with the value and significance of log likelihood tests for λ = 0 (phylogenetic independence) and λ = 1 (complete dependence). Significance is indicated by: *P < 0.05, **P < 0.01, ***P < 0.001
 λ (95% CI)ln likln lik (λ = 0)ln lik (λ = 1)
AOO10.234 (0.085–0.439)−205.1−211.5***−256.6***
AOO100.0 (0–0.166)−213.0−213.0−276.3***
ResidD1−100.446 (0.162–0.695)−577.9−583.5***−608.3***
ResidD10−1000.167 (0–0.397)−553.0−556.7**−606.3***

univariate analyses

Given that the distributional variables AOO1, ResidD1−10 and ResidD10−100 all showed a degree of phylogenetic dependence, we carried out two sets of univariate analyses: cross-species and phylogenetically informed analyses, and compared the results. The first set of analyses used general linear models to test the effect of plant characteristics on AOO1, D1−10 (while controlling for AOO1) and D10−100 (while controlling for AOO10). Ryan–Einot–Gabriel–Welsch (REGW) post-hoc tests were performed on ResidDB to test for differences between categories where the categorical variable was significant (Quinn & Keough 2002).

phylogenetically informed analyses

Phylogenetically informed analyses were carried out with phylogenetic regression (phylo.glm, version 1.03; Grafen 1989), using the phylogeny given in Appendix S1. This program was specifically selected because it can accommodate multiple predictor variables and allows for the simultaneous inclusion of both continuous and categorical data (Grafen 1989). Path segment lengths were specified according to figure 2 in Grafen (1989). Phylogenetic regression does not produce trustworthy measures of the fit of the data (R2; Grafen 1989), so only F-values and their significance are reported in the results for both the cross-species and phylogenetically informed analyses.

AOO remained a significant predictor of fractal dimension when phylogeny was taken into account. When included together in an analysis, AOO1 was a significant predictor of D1−10 (F1,222 = 8.98, P = 0.003) whereas D10−100 was not (F1,222 = 0.03, P = 0.869), and AOO10 was a significant predictor of D10−100 (F1,223 = 7.45, P = 0.007), just as in the cross-species analysis. Consequently, AOO (at the appropriate scale) was included as a covariate in phylogenetic analyses of trait effect on fractal dimension, as with the cross-species analyses above.

multivariate analyses

The traits under consideration are interrelated and can potentially interact in affecting distribution patterns. Considering all variables together dramatically reduced the sample size because missing data were compounded. We therefore considered subsets of conceptually related variables in separate multivariate analyses (Table 1). We did not consider habitat in these analyses due to the large number of factors present. With groups of derived variables, the variable included in the multivariate analysis was the one that was the most significant predictor in univariate analyses, or, in the absence of variables being significant, was the variable with the greatest number of categories. All possible combinations of the variables (with and without interactions) were considered, up to models with two main effects and a two-way interaction. The most parsimonious models in each group were assessed using Akaike's Information Criterion (AIC), and models with a difference of less than 4.0 from the minimum AIC were considered to have reasonably good support (Burnham & Anderson 1998). We undertook only cross-species multivariate analyses, because model selection was not possible with the output from phylogenetic regression. However, the results of multivariate analyses were interpreted with respect to differences between cross-species and phylogenetically informed univariate analyses.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

correlations between variables

We considered 18 attributes of British rare and scarce plants relating to life history, morphology and changes in distribution (Table 1). There were significant associations between only four pairs of attributes after Bonferonni correction (Pcorr). Not surprisingly, the presence of a pollen vector was significantly related to the mode of fertilization (Kendall's τb = −0.535, Pcorr < 0.001). This result confirms our predictions, as cross-fertilized species require a pollen vector, whereas many self-fertilized species neither possess nor require one. For the same reason, the type of pollen vector was also related to mode of fertilization (Fisher's exact test, Pcorr < 0.001). However, there was no significant difference in fertilization mode between insect-pollinated and wind/water-pollinated taxa (Kendall's τb = 0.012, Pcorr > 0.10). There was a positive correlation between the number of seeds produced per flower and the number produced per plant (Kendall's τb = 0.400, Pcorr = 0.036) and a negative correlation between the number of seeds produced per flower and log-transformed seed mass (Kruskal–Wallis H = 30.850, Pcorr < 0.001). This ecological trade-off between few large seeds or many small ones has been widely noted and may underpin the trade-off between colonization and competitive ability in plants (Werner & Platt 1976; Tilman 1994).

distributional analyses

The relationships between plant characteristics and distributions (AOO and fractal dimension at two scales) are given in Table 3. Significance levels are reported without Bonferroni correction, because they tested specific a priori hypotheses (given in the Introduction). Of the 18 variables tested, two were significantly related to AOO1, six to D1−10 and five to D10−100 (Table 3). All of the significant relationships (P < 0.05) were in the predicted direction, except for variables relating to pollination vector. The results of the phylogenetically informed analysis were generally similar to the cross-species analysis.

Table 3.  Regression coefficients for univariate cross-species (CS) and phylogenetically informed (PI) relationships of plant traits with AOO1 (log-transformed area of occupancy at 1-km scale), D1−10 and D10−100 (box counting fractal dimensions between 1- and 10-km scales and 10- and 100-km scales, respectively). A dash indicates P > 0.10, +P < 0.10, *P < 0.05, **P < 0.01, ***P < 0.001. C, coefficients not shown for significant categorical variables with more than one category
  AOO1 D 1−10 D 10−100
CSPICSPICSPI
1Raunkiaer life-formC+
2Woodiness 0.026+
3Local abundance0.224***0.224*** 0.101*** 0.101*** 0.038+
4Propagation mode −0.031+ −0.036*
5aFertilization mode
5bObligate outcrossing 0.077+ 0.079+ 0.081* 0.074*
6aType of pollen vectorC**
6bPresence of pollen vector
6cInsect pollinated 0.077***
7aSeed mass
7bSeed production (per flower)
7cSeed production (per plant)
8aDiaspore dispersalC+C+C+C+
8bDispersal (specialized vs. unspecialized) −0.101** −0.101**
8cDispersal (long vs. short) −0.077* −0.080* −0.049+ −0.050+
9Change index 0.032* 0.034* 0.025+ 0.024+
10Dynamism index −0.298*** −0.304*** −0.214*** −0.205***
11HabitatC*C***C*C**C**

Multivariate analyses indicated that in most cases (12 of 15) the best approximating model contained only a single main effect (Table 4). In one case, two plant traits had an additive effect in the best model and in two cases two traits interacted to affect plant distributions. It was not possible, with the currently available software, to conduct model selection with multivariate analyses using a phylogenetically informed approach. However, the similarities between results from the cross-species and phylogenetically informed univariate analyses (Table 3) suggested that our approach would be a good approximation to results from an analysis with the appropriate level of phylogenetic correction.

Table 4.  Best approximating models from each subset of multivariate models. Variables were selected for inclusion from each multivariate group as described in the text. All possible combinations of the variables in each subset, up to two main effects and their interaction, were considered. ΔAIC, the difference between the model and the best approximating model in each subset of analyses (i.e. that with the minimum AIC). All models where ΔAIC < 4.0 are shown. Significance is as indicated in Table 3, with the absence of any symbol indicating P > 0.1
Dependent variableMultivariate groupPredictor variables in the modelRank and description of models with good support. (Significance of terms in the model is indicated.)ΔAIC
AOO11Raunkiaer life-form, woodiness1. Woodiness***0
2Local abundance, propagation mode1. Local abundance***0
3Local abundance, fertilization mode, pollination mode1. Fertilization mode0
  Local abundance*** 
  Interaction 
4Seed mass, seed production (per flower), dispersal mode1. Seed production (per flower)+0
  2. Seed production (per flower)+1.79
  Dispersal mode 
  Interaction* 
5Local abundance, change, dynamism1. Dynamism+0
  Local abundance*** 
  Interaction*** 
D 1−10 1Raunkiaer life-form, woodiness1. Woodiness+0
2Local abundance, propagation mode1. Local abundance***0
3Local abundance, obligate outcrossing, pollination mode1. Local abundance***0
  2. Obligate outcrossing3.62
  Local abundance*** 
4Seed mass, seed production (per flower), dispersal mode1. Seed production (per flower)0
  2. Seed mass1.04
5Local abundance, change, dynamism1. Dynamism***0
  Local abundance*** 
  2. Dynamism2.57
  Local abundance* 
  Interaction 
D 10−100 1Raunkiaer life-form, woodiness1. Woodiness0
2Local abundance, propagation mode1. Local abundance0
  2. Propagation mode3.93
3Local abundance, obligate outcrossing, insect- vs. wind-pollinated1. Obligate outcrossing0
  2. Insect- vs. wind-pollinated0.06
  3. Obligate outcrossing*1.11
  Insect- vs. wind-pollinated 
  Interaction* 
  4. Obligate outcrossing*2.88
  Insect- vs. wind-pollinated 
4Seed mass, seed production (per flower), dispersal mode1. Seed production (per flower)0
  2. Seed mass0.08
5Local abundance, change, dynamism1. Dynamism*0
Local abundance and habitat

Local abundance, an index of local field density or dominance (assessed at approximately 10–100 m resolution) was positively associated with larger range sizes (high AOO1) and relatively high levels of aggregation at scales of 1–10 km (high D1−10) (Fig. 2a,c). In the univariate tests, habitat was the only other variable significantly related to AOO1 (Fig. 2b). The post-hoc REGW test showed that taxa associated with montane and water margin habitats had lower AOO1 than those preferring freshwater habitats. Habitat was also an important explanatory variable for distribution patterns (D1−10 and D10−100). Post-hoc REGW tests on ResidD1−10 were unable to detect differences between habitats (Fig. 2e). The differences in ResidD10−100 are shown in Fig. 2h and, as expected, the distribution pattern of a species tended to mirror the apparent distribution pattern of its habitat.

image

Figure 2. Boxplots of categorical variables significantly related to the distribution patterns of rare and scarce British plants. Comparisons are shown for: AOO1 (log-transformed area of occupancy at 1-km scale) with (a) local abundance and (b) habitat; ResidD1−10 (standardized residual of box-counting fractal dimension between 1- and 10-km scales against AOO1) with (c) local abundance, (d) dispersal mode (significant comparisons based on a priori hypotheses are shown) and (e) habitat; ResidD10−100 (standardized residual of box-counting fractal dimension between 10- and 100-km scales against AOO10) with (f) obligate outcrossing, (g) pollen vector (significant comparisons based on a priori hypotheses are shown) and (h) habitat. The dark line is the median, the box encloses the interquartile range and the whiskers show the full range, with outliers shown as circles. Sample sizes are given above the boxplots. Where attributes resulting from redefinition of the original attributes were tested and found to be significant, they are indicated as *P < 0.05, **P < 0.01, ***P < 0.001. Where there were differences between categories, AOO1 or ResidDB was tested with post-hoc Ryan–Einot–Gabriel–Welsch (REGW) tests and groups that did not significantly differ are indicated by horizontal lines above the graph. Explanations of abbreviations for habitat type are given in Table 1.

Download figure to PowerPoint

Life-form, woodiness and propagation

Life-form and woodiness are both potential surrogates for longevity, but neither was an important explanatory variable in the univariate tests (Table 3). They did not significantly interact. Propagation mode had only a weak, or no, effect on distribution patterns. It was conceptually related to local abundance, but the latter variable explained most of the variation in multivariate models (Table 4).

Mode of fertilization and pollination vector

Several variables related to pollination were associated with fractal dimension at one or both scales. Whereas the three-state categorical variable of fertilization mode did not significantly predict DB at either scale, a binary analysis comparing obligate outcrossers with all other species demonstrated that obligate outcrossers had relatively more aggregated distributions than other species, at a resolution of 10–100 km (Fig. 2f). Pollination vector, as a four-state categorical variable, had a strong effect on D10−100 (Fig. 2g), but surprisingly little effect on D1−10. A comparison of species with and without a pollination vector showed no significant effects on either DB at either scale. However, of those species with a pollination vector, insect-pollinated species showed significantly more aggregated coarse-scale distributions (higher D10−100) than those using wind or water as a pollen vector (Fig. 2g). These effects potentially interact with local abundance, but considering all variables together, the majority of variation in AOO1 and D10−100 is explained by local abundance (Table 4).

Diaspore dispersal, number of seeds and seed mass

In the univariate tests, seed number and seed mass had no significant effect on any of the indices of distribution pattern, whereas the mode of diaspore dispersal had marginally significant effects on aggregation at both scales (0.1 > P > 0.05). Notably, taxa with long-distance dispersal mechanisms (via wind, water or vertebrates) had significantly more scattered distributions at fine scales (low D1−10) compared with taxa with short-distance dispersal mechanisms (Fig. 2d). The same was true for taxa having specialized dispersal compared with those without (Fig. 2d). Although these variables are conceptually related, and are subject to evolutionary and ecological trade-offs, the best approximating models in the multivariate analysis were always non-significant single-factor models involving the number of seeds per flower (Table 4).

Change and dynamism

The contemporary distribution patterns (D1−10 and D10−100) of rare and scarce British plants were significantly related to the amount of change and turnover (dynamism) in the number of occupied 10-km squares that a species had exhibited over the past half century. Taxa with increasing range sizes had relatively more aggregated distributions compared with decreasing taxa, and those with highly dynamic ranges had relatively scattered distributions compared with taxa with more stable ranges. The combined effect is that the most scattered distributions are associated with those taxa that are declining and have highly dynamic ranges (Fig. 3). Dynamism was always present in the most parsimonious model in the multivariate analyses; on its own to predict D10−100, as a additive effect with local abundance to predict D1−10 and interacting with local abundance in parsimonious models to predict AOO1 (Table 4). Specifically, dynamism had a weak effect on AOO1 for locally scattered taxa and a strong effect for locally frequent and locally abundant taxa.

image

Figure 3. Plane of best fit for fine-scale fractal dimension (D1−10) of 292 taxa of rare and scarce British plants against the change index and dynamism index (as described in the text). The response variable is the residual of D1−10 against AOO1. The relationship between the change and dynamism indices and the number of reported colonizations and extinctions at the 10-km scale (Biological Records Centre data) is also shown. The plane of best fit is very similar to that for D10−100.

Download figure to PowerPoint

Our findings suggests that the change and dynamism indices reflect two aspects of a more general pattern, as they are both derived from the measured rates of colonization and extinction. The change index is affected positively by colonization rate, but negatively by extinction rate, whereas the dynamism index is affected positively by both. Thus, the observed positive association of DB with change and its negative association with dynamism suggest a negative association between DB and extinction, which could underlie the observed effects on both indices. To examine the interrelationships in more detail we compared models that included the primary variables (colonization and extinction) singly, together and with an interaction term, and compared them with models composed of the derived indices (Table 5). The most parsimonious model for D1−10, assessed with AIC, included both colonization and extinction rate (with extinction being the stronger effect), although the model with extinction rate alone had almost as good support as the best model (Table 5). The most parsimonious model for D10−100 was the one that included extinction rate only. Almost without exception, colonization and extinction rates produced better models than those involving the derived indices.

Table 5.  Regression coefficients for models including interactions between variables measuring change in species distribution patterns. Patterns are quantified using the fractal dimension of point distributions at linear scales of 1–10 km (D1−10) and 10–100 km (D10−100). Log-transformed area of occupancy at the appropriate scale (AOO1 or AOO10) was used as a covariate in all models. Details of the variables are given in the text. CH, change index; DYN, dynamism index; COL, proportional rate of colonization; EXT, proportional rate of extinction; AIC, Akaike's information criterion, ΔAIC, the difference between the model and the best approximating model in each set of analyses (minimum AIC). Significance is indicated as in Table 3. Where ΔAIC < 2 there is strong support for the model
ModelCHDYNCH × DYNCOLEXTCOL × EXTAICΔAIC
D 1−10
 1 0.032*     −144.9623.18
 2 −0.298***    −165.49 2.65
 3 0.039**−0.314***    −166.30 1.84
 4 0.087−0.319***−0.081   −164.29 3.85
 5   −0.208***  −160.24 7.91
 6    −0.266*** −166.54 1.60
 7   −0.127*−0.204*** −168.14 0.00
 8   −0.070−0.142−0.140−167.48 0.67
D 10−100
 1 0.025+     −207.0213.89
 2 −0.214***    −220.39 6.06
 3 0.031*−0.227***    −219.1610.33
 4−0.074−0.215***   0.177+   −219.41 7.31
 5   −0.109*  −215.3811.33
 6    −0.217*** −229.65 0.00
 7   −0.029−0.204*** −225.81 2.90
 8      0.020−0.150−0.121−224.91 3.27

Although the change index is standardized (with a mean of zero) over all British vascular plant species, the rare and scarce plants in our data set had a marginally significant tendency towards negative change values (mean = −0.80, t291 = −1.866, P = 0.063). This suggests that rare and scarce plants have declined in AOO relative to the remainder of the British flora.

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

plant traits and national abundance

This study tested the effect of 18 plant traits on AAO at 1 km resolution, but only two were significantly related to AOO. The reason that more were not related may have been because we restricted our analysis to rare and scarce species (the taxa with the best documented distributions at 1-km resolution), so the range of AOO was not as great as in comparisons of rare and common species (see Introduction). It may also be that there are some factors affecting the AOO of these rare and scarce taxa that are specific to this subset of all plants, e.g. habitat management specifically for the plants under consideration. Despite this, it is surprising that so few of these traits correlate with AOO. The only two variables considered that had significant effects on AOO were habitat preference and local abundance. The effect of habitat is a simple one: species that occupy widespread habitats have the potential to occupy more area than species that require rare habitat types, and habitat availability is subject to strong anthropogenic influences. The relationship between typical abundance and AOO demonstrates that plant taxa that are typically locally dominant occupy more 1-km squares in Britain than do species that locally grow more sparsely. This provides further confirmation of the relationship between regional distribution and local abundance (e.g. Bock & Ricklefs 1983; Blackburn et al. 1997), although a wide range of mechanisms have been proposed to explain the pattern (reviewed by Gaston et al. 2000; Holt et al. 2002).

plant traits and distribution patterns

This study is one of the largest to examine the factors influencing species spatial distribution patterns. It extends the work of Quinn et al. (1994) by investigating additional species and predictor variables, and includes multivariate analyses to test for interactions between variables. Quinn et al. (1994) investigated the relationship of five plant traits to an index of aggregation: three showed significant relationships (habitat, dispersal mode and fertilization mode) and two did not (life history and seed size). Our work increased the number of taxa under consideration (391 compared with 139) and used fractal dimension (with AOO as a covariate) as the index of aggregation, yet the results were similar: there were significant (P < 0.05) or marginally significant (P < 0.10) results for habitat, dispersal mode and fertilization mode, and non-significant results for Raunkiaer life-form and seed mass.

In total, we considered 18 plant traits which showed variation in their degree of association with distribution patterns (Table 3). Although some traits were conceptually related, most did not significantly interact to affect distribution patterns (Table 4). Obligate outcrossing plants, i.e. dioecious or self-incompatible plants, have significantly more aggregated distributions than plants that can self-fertilize, as predicted, which we consider to be most likely due to the ability of predominantly self-fertilized species to establish and/or persist in isolated populations (Baker 1955; Quinn et al. 1994; Barrett et al. 1996). Insect-pollinated plants also have significantly more aggregated distributions than wind- or water-pollinated plants. This was contrary to our expectations, but little is known of the distance decay curves of effective pollen transfer via wind and water or insects. The relationship was only significant at coarse scales (10–100 km), suggesting that at distances greater than 10 km wind pollination may be more effective than insect pollination.

Seed production and seed mass are subject to complex trade-offs with dispersal mode (Thompson et al. 2002), but in our analysis only dispersal mode had a significant effect on distribution pattern. Specialized modes of dispersal and features that allowed long-distance dispersal, such as epizoochory and wind-dispersed seeds, allow individuals to establish far from their parent and so cause more scattered distributions, at least at fine scales of 1–10 km. We did not consider the effect of the persistence of seed in the seed bank, which has previously been shown to have no significant effect on distribution patterns, even though it may allow ‘dispersal’ through time (Thompson & Hodgson 1996).

The relationship between typical local abundance (which reflects population density at very fine spatial scales, ∼10–100 m) and measures of distribution pattern at coarser scales (1–10 km) confirms that coarse-scale occurrence patterns can help to predict local abundance (He & Gaston 2000; Kunin et al. 2000). Note, however, that the relationship is less strong at coarser scales (10–100 km), indicating that species distributional patterns may generalize across scales, but they cannot be extrapolated indefinitely. In a survey of 16 of the species considered in the current study, Hartley et al. (2004) also found strong cross-scale correlations in spatial patterning, but noted that these relationships weaken as the range of scales becomes increasingly wide and may break down abruptly across particular scales. In this more extensive species set, however, it is apparent that the relationship between spatial patterns across scales does not break down completely, allowing some residual predictive power between local and regional scaling domains.

Habitat preference was the only variable to have a significant effect on DB at both scales, as well as on AOO1. Taxa with distributions having relatively low fractal dimensions occur in habitats that are patchy at national and local scales, e.g. coastal habitats and fens. Those having relatively high fractal dimensions occur in habitats that are locally prevalent but restricted nationally, e.g. calcareous habitats or heaths, bogs and moors (Stewart et al. 1994).

Although plant traits could potentially act together to affect distribution patterns, we did not find evidence to support this in most of our multivariate analyses. This is probably partly due to colinearity, so the variation explained by one trait (e.g. life-form) is equally well explained by another (e.g. woodiness). The power of the tests, given the variation in the data, may also have been a factor.

Although some of the traits that we considered were associated with distribution patterns, there are several possible reasons why more of the traits were not associated. Although we consider that processes will be affecting the distribution patterns of rare and common species similarly, it is possible that rare and scarce plants are subject to particular processes by fact of their rarity. Where we derived our predictions from published studies, they were mainly comparing rare and common taxa and restricting the analyses to relatively rare taxa may have limited the strength of relationships. However, it did allow us to consider fine-scale patterns (at scales below 10 km) as a result of the availability of data. In addition, distributions with small AOOs have greater potential variation in fractal dimension than more space-filling distributions, i.e. with a higher AOO. The generality of our results could be tested by examining a greater range of taxa, e.g. the full range of British plants (Preston et al. 2002), although data for commoner species would only be available for coarser scales (> 10-km resolution). Because the calculation of box-counting fractal dimensions requires aggregating the data to still coarser scales (e.g. 100-km resolution), and because many of the commoner species would occupy all the UK grid cells at this scale, the pattern information would be limited in its usefulness. Moreover, it is possible, as suggested by Thompson & Hodgson (1996), that stronger relationships would be found between trait data and distributional patterns in landscapes subject to less anthropogenic manipulation.

distribution patterns and changes in distribution

Although spatial distribution patterns are poorly explained by life history or morphological traits, they are strongly associated with temporal range dynamics. We measured two aspects of changing distributions using the change index standardized over the whole flora of Britain (Telfer et al. 2002) and an index of dynamism (or spatio-temporal turnover). The change index takes account of differences in recording effort between survey periods (Telfer et al. 2002), and therefore it is an unbiased estimate of change in distributions, but dynamism does not take account of such differences (Preston et al. 2003), and so may show more bias. The length of the surveys may also cause change and, particularly, dynamism to be underestimated, so our analyses will tend to be conservative. However, both indices showed strong effects on the distribution patterns of rare and scarce British plants (Table 3) and they significantly interact (Fig. 3, Table 5), although local abundance does explain much of the variation when included in multivariate models (Table 4). Taxa with relatively aggregated distributions tend to be increasing in abundance and have stable ranges. Those with relatively scattered distributions tend to be declining and have dynamic ranges. Those that are more locally abundant show stronger effects of dynamism on fine-scale aggregation and the two factors interact to affect AOO (Table 4). The indices of change and dynamism are not strongly correlated, but including both terms in a single model strengthens the effect of both, as each explains variation that obscures the other's effect (Model 3 in Table 5). The two indices comprise the same basic components: colonization rate and extinction rate. The rate of extinction was shown to have a strong effect on fractal dimension at both spatial scales (explaining as much of the variation in DB at each scale as the change and dynamism indices combined), whereas the rate of colonization had very little effect (Models 5–8 in Table 5).

Although a higher extinction rate was related to a more scattered distribution, it is difficult to disentangle cause and effect in this relationship. High rates of local extinctions could cause fragmentation of species ranges, particularly if extinction events are discrete and not spatially autocorrelated (Hartley & Kunin 2003; but see Channell & Lomolino 2000). Alternatively, a scattered distribution may itself increase the rate of extinction, as widely dispersed populations provide little opportunity for rescue effects (Brown & Kodric-Brown 1977) or recolonization. Furthermore, given the correlation of spatial pattern across scales (Kunin 1998; Hartley et al. 2004), and the association between low D1−10 and low local abundance reported here, it is likely that, for a given AOO1, species with low D1−10 have lower AOO at subkilometre resolutions, and ultimately smaller local population sizes, thus leaving them at greater risk of extinction due to the problems of demographic stochasticity and inbreeding depression (Caughley 1994). Taken together, these arguments strongly suggest the possibility of an ‘extinction vortex’ (Gilpin & Soulé 1986), in which extinction and fragmentation are mutually reinforcing. A similar relationship was also found in British and Belgian butterflies by Wilson et al. (2004), who further demonstrated that scattered distributions in the past tended to predict high rates of subsequent local extinction. These results suggest that conservation priority should be focused on species with scattered ranges as these appear most likely to decline (cf. Pearman 1997). Our results could shed some light on the ‘trajectories to extinction’ exhibited by rare plant species. Schoenwald-Cox & Buechner (1991) suggested that populations could in principle decline in strikingly different ways: at one extreme abundance could decrease across all subpopulations (with no effect on site occupancy), or at the opposite extreme it might lose some subpopulations (sites) completely, with no change in abundance at the remaining sites. If we replace ‘individuals’ with ‘fine-scale cells occupied within a coarse-scale cell’, this then provides a context within which to interpret our findings. Our results suggest a degree of hysteresis in range dynamics such that declines create a diffuse structure of relictual subpopulations. Subpopulations (individuals or fine-scale cells) seem to disappear in a fairly spatially random manner, such that the coarse-scale structure persists, whereas, when distributions expand, subpopulations increase in local density and new subpopulations form close to existing ones.

phylogenetically informed analysis

It used to be common practice to perform interspecific comparative analyses without including phylogenetic information, thereby treating each species as an independent data point. Such analyses can be misleading, as traits may be associated with one another by inheritance rather than due to adaptive evolution (Felsenstein 1985) and so cross-species analyses exaggerate the statistical power of the analysis (Harvey & Pagel 1991).

However, the current use of phylogenetically informed analyses is not consistent between authors. Some contend that cross-species and phylogenetically informed analyses are complementary methods that address different questions (Wright et al. 2000; Murray et al. 2002; Hamilton et al. 2005). Others state that, because these two approaches make such fundamentally different assumptions, it is impossible for both to be correct in any single analysis (Freckleton et al. 2002).

Phylogenetically informed analyses can be used simply for statistical correction, to account for the lack of statistical independence in traits that show phylogenetic dependence (i.e. they are autocorrelated). Similar approaches are used in spatial statistics, which is also subject to problems of lack of statistical independence. Consider the correlation between two variables, one of which is spatially autocorrelated (i.e. has an autocorrelation function greater than zero) and one of which is not (i.e. its autocorrelation function is zero). There would be no need to correct for the lack of statistical independence in the data set because the effective degrees of freedom are affected by the product of the autocorrelation functions (Clifford et al. 1989), which will always equal zero when one variable is not autocorrelated. Similarly, when analysing a relationship between two species traits, one of which is phylogenetically dependent (i.e. its autocorrelation function is greater than zero) and one of which is not (i.e. its autocorrelation function is equal to zero), a cross-species analysis is statistically valid. Rheindt et al. (2004) have argued otherwise. In their analysis they took account of phylogenetic information only for the phylogenetically autocorrelated trait, by calculating phylogenetically independent contrasts (PICs) from the best available phylogenetic tree (for the autocorrelated trait) and PICs from an evolutionary neutral star phylogeny (for the independent trait). However, work in progress (by W.E. Kunin) on simulated datasets suggests that simple (uncorrected) analyses are more appropriate than either this or more conventional phylogenetic approaches for data of this kind.

The ability to assess phylogenetic dependence (e.g. the value of λ in Table 2; Freckleton et al. 2002) allows the appropriate analysis to be selected. We found stronger phylogenetic dependence in fine-scale distributional variables and considered that a degree of phylogenetic correction was appropriate (Freckleton et al. 2002). The format of our analysis did not allow this to be performed using currently available software, so we undertook and compared the results of both phylogenetically informed and cross-species univariate analyses.

In general, we found slightly weaker relationships in the phylogenetically informed analyses, due partly to slightly smaller sample sizes (number of taxa in cross-species analyses compared with PICs in phylogenetically informed analyses). The most notable difference between the two types of analyses was the relationship between D10−100 and the presence of an insect pollination vector, which completely disappeared in the phylogenetically informed analysis. This is because pollination vector is a strongly conserved trait, so the present-day relationship is a result of a small number of correlated divergences between fractal dimension and pollination vector (Hamilton et al. 2005), i.e. the statistical power of the test is weak.

Conclusions

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

The findings of this study are relatively robust to methodology. We provide clear evidence that the spatial pattern of plant species distributions is influenced by a range of ecological traits, in particular those concerning dispersal, pollination, habitat and local density. Most strikingly, we find that both the spatio-temporal dynamism and net change in distributions are strongly associated with distribution parameters, with potential implications for conservation effort.

Caughley (1994) identified two paradigms in conservation biology: the small population paradigm and the declining population paradigm. The results presented here suggest that species with highly fragmented distribution patterns (measured at resolutions of a kilometre or so) are likely to qualify on both counts, thus placing them at the highest risk of extinction. The relationship between extinction and spatial pattern suggests that, in the absence of better information, static distributional patterns can provide clues as to the recent history, and indeed the likely future change, of populations. Clearly, the analysis of species distribution patterns has the potential to provide useful insights for conservation and management.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

This work was supported by NERC grant GR3/11916. Thanks to Peter Mayhew for valuable discussions regarding phylogenetic regression and to Linley Jesson, Marc Cadotte and an anonymous referee for comments on the manuscript.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information
  • Allison, T.D. (1990) Pollen production and plant density affect pollination and seed production in Taxus canadensis. Ecology, 71, 516522.
  • Andrewartha, H.G. & Birch, L.C. (1954) The Distribution and Abundance of Animals. The University of Chicago Press, Chicago.
  • Angiosperm Phylogeny Group (1998) An ordinal classification for the families of flowering plants. Annals of the Missouri Botanical Garden, 85, 531553.
  • Baker, H.G. (1955) Self-compatibility and establishment after ‘long-distance’ dispersal. Evolution, 9, 347349.
  • Barrett, S.C.H., Harder, L.D. & Worley, A.C. (1996) The comparative biology of pollination and mating in flowering plants. Philosophical Transactions of the Royal Society of London B, 351, 12711280.
  • Blackburn, T.M., Gaston, K.J., Quinn, R.M., Arnold, H. & Gregory, R.D. (1997) Of mice and wrens: the relation between abundance and geographic range size in British mammals and birds. Philosophical Transactions of the Royal Society of London B, 352, 419427.
  • Bock, C.E. & Ricklefs, R.E. (1983) Range size and local abundance of some North American songbirds: a positive correlation. American Naturalist, 122, 295299.
  • Brown, J.H. (1995) Macroecology. University of Chicago Press, Chicago.
  • Brown, J.H. & Kodric-Brown, A. (1977) Turnover rates in insular biogeography: effect of immigration on extinction. Ecology, 58, 445449.
  • Burnham, K.P. & Anderson, D.R. (1998) Model Selection and Inference. A Practical Information-Theoretic Approach. Springer-Verlag, New York.
  • Cadotte, M.W. & Lovett-Doust, J. (2002) Ecological and taxonomic differences between rare and common plants of southwestern Ontario. Ecoscience, 9, 397406.
  • Cadotte, M.W., Murray, B.R. & Lovett-Doust, J. (2006) Ecological patterns and biological invasions: using regional species inventories in macroecology. Biological Invasions, doi: 10.1007/s10530-005-3839-4.
  • Caughley, G. (1994) Directions in conservation biology. Journal of Animal Ecology, 63, 215244.
  • Channell, R. & Lomolino, M.V. (2000) Trajectories to extinction: spatial dynamics of the contraction of geographical ranges. Journal of Biogeography, 27, 169179.
  • Clapham, A.R., Tutin, T.G. & Warburg, E.F. (1952) Flora of the British Isles. Cambridge University Press, Cambridge.
  • Clifford, P., Richardson, S. & Hémon, D. (1989) Assessing the significance of the correlation between two spatial processes. Biometrics, 45, 123134.
  • Dale, M.R.T. (1999) Spatial Pattern Analysis in Plant Ecology. Cambridge University Press, Cambridge.
  • Edwards, W. & Westoby, M. (1996) Reserve mass and dispersal investment in relation to geographic range of plant species: phylogenetically independent contrasts. Journal of Biogeography, 23, 329338.
  • Felsenstein, J. (1985) Phylogenies and the comparative method. American Naturalist, 125, 115.
  • Fitter, A.H. & Peat, H.J. (1994) The ecological flora database. Journal of Ecology, 82, 415425.
  • Freckleton, R.P., Harvey, P.H. & Pagel, M. (2002) Phylogenetic analysis and comparative data: a test and review of evidence. American Naturalist, 160, 712726.
  • Garland, T., Harvey, P.H. & Ives, A.R. (1992) Procedures for the analysis of comparative data using phylogenetically independent contrasts. Systematic Biology, 41, 1832.
  • Gaston, K.J. & Blackburn, T.M. (2000) Pattern and Process in Macroecology. Blackwell Science, Oxford.
  • Gaston, K.J., Blackburn, T.M., Greenwood, J.J.D., Gregory, R.D., Quinn, R.M. & Lawton, J.H. (2000) Abundance–occupancy relationships. Journal of Applied Ecology, 37, S39S59.
  • Gaston, K.J. & Kunin, W.E. (1997) Rare-common differences: an overview. The Biology of Rarity (eds W.E. Kunin & K.J. Gaston), pp. 1229. Chapman & Hall, London.
  • Gilpin, M.E. & Soulé, M.E. (1986) Minimum viable populations: process of species extinctions. Conservation Biology. the Science of Scarcity and Diversity (ed. M.E. Soulé), pp. 1934. Sinauer Associates, Sunderland, MA.
  • Grafen, A. (1989) The phylogenetic regression. Philosophical Transactions of the Royal Society of London B, 326, 119157.
  • Halley, J.M., Hartley, S., Kallimanis, A.S., Kunin, W.E., Lennon, J.J. & Sgardelis, S.P. (2004) Uses and abuses of fractal methodology in ecology. Ecology Letters, 7, 254271.
  • Hamilton, M.A., Murray, B.R., Cadotte, M.W., Hose, G.C., Baker, A.C., Harris, C.J. & Licari, D. (2005) Life-history correlates of plant invasiveness at regional and continental scales. Ecology Letters, 8, 10661074.
  • Harper, K.T. (1979) Some reproductive and life history characteristics of rare plants and implications for management. Great Basin Naturalist Memoirs, 3, 129137.
  • Hartley, S. (1998) A positive relationship between local abundance and regional occupancy is almost inevitable (but not all positive relationships are the same). Journal of Animal Ecology, 67, 992994.
  • Hartley, S. & Kunin, W.E. (2003) Scale dependency of rarity, extinction risk and conservation priority. Conservation Biology, 17, 15591570.
  • Hartley, S., Kunin, W.E., Lennon, J.J. & Pocock, M.J.O. (2004) Coherence and discontinuity in the scaling of species distribution patterns. Proceedings of the Royal Society Biological Sciences B, 271, 8188.
  • Harvey, P.H. & Pagel, M.D. (1991) The Comparative Method in Evolutionary Biology. Oxford University Press, Oxford.
  • He, F. & Gaston, K.J. (2000) Estimating species abundance from occurrence. American Naturalist, 156, 553559.
  • Hegde, S.G. & Ellstrand, N.C. (1999) Life history differences between rare and common flowering plant species of California and the British Isles. International Journal of Plant Sciences, 160, 10831091.
  • Holt, A.R., Gaston, K.J. & He, F. (2002) Occupancy–abundance relationships and spatial distribution: a review. Basic and Applied Ecology, 3, 113.
  • Kelly, C.K. (1996) Identifying plant functional types using floristic data bases: ecological correlates of plant range size. Journal of Vegetation Science, 7, 417424.
  • Kelly, C.K. & Woodward, F.I. (1996) Ecological correlates of plant range size: taxonomies and phylogenies in the study of plant commonness and rarity in Great Britain. Philosophical Transactions of the Royal Society of London B, 351, 12611269.
  • Kessler, M. (2002) Environmental patterns and ecological correlates of range size among bromeliad communities of Andean forests in Bolivia. Botanical Review, 68, 100127.
  • Klinkenberg, B. & Goodchild, M.F. (1992) The fractal properties of topography: a comparison of methods. Earth Surface Processes and Landforms, 17, 217234.
  • Klinkhamer, P.G.L., de Jong, T.J. & de Bruyn, G.-J. (1989) Plant size and pollinator visitation in Cynoglossum officinale. Oikos, 54, 201204.
  • Kunin, W.E. (1992) Density and reproductive success in wild populations of Diplotaxis erucoides (Brassicaceae). Oecologia, 91, 129133.
  • Kunin, W.E. (1993) Sex and the single mustard: population density and pollinator behavior effects on seed-set. Ecology, 74, 21452160.
  • Kunin, W.E. (1997) Population size and density effects in pollination: pollinator foraging and plant reproductive success in experimental arrays of Brassica kaber. Journal of Ecology, 85, 225234.
  • Kunin, W.E. (1998) Extrapolating species abundance across spatial scales. Science, 281, 15131515.
  • Kunin, W.E. & Gaston, K.J. (1993) The biology of rarity: patterns, causes and consequences. Trends in Ecology and Evolution, 8, 298301.
  • Kunin, W.E. & Gaston, K.J. (1997) The Biology of Rarity. Chapman & Hall, London.
  • Kunin, W.E., Hartley, S. & Lennon, J.J. (2000) Scaling down: on the challenge of estimating abundance from occurrence patterns. American Naturalist, 156, 560566.
  • Lloyd, K.M., Lee, W.G. & Wilson, J.B. (2002) Growth and reproduction of New Zealand Acaena (Rosaceae) species in relation to rarity and commonness. New Zealand Journal of Ecology, 26, 149160.
  • MacArthur, R.H. (1972) Geographical Ecology. Harper & Row, New York.
  • McAlpine, R.S. & Wotton, B.M. (1993) The use of fractal dimension to improve wildland fire perimeter predictions. Canadian Journal of Forest Research, 23, 10731077.
  • Murray, B.R., Thrall, P.H. & Lepschi, B.J. (2002) Relating species rarity to life history in plants of eastern Australia. Evolutionary Ecology Research, 4, 937950.
  • Ostling, A., Harte, J., Green, J.L. & Kinzig, A.P. (2003) A community-level fractal property produces power-law species–area relationships. Oikos, 103, 218224.
  • Pagel, M. (1999) Inferring the historical patterns of biological evolution. Nature, 401, 877884.
  • Pearman, D.A. (1997) Towards a new definition of rare and scarce plants. Watsonia, 21, 231251.
  • Peat, H.J. & Fitter, A.H. (1994) Comparative analyses of ecological characteristics of British angiosperms. Biological Reviews, 69, 95115.
  • Peitgen, H.-O., Jürgens, H. & Saupe, D. (1992) Fractals for the Classroom. Part One: Introduction to Fractals and Chaos. Springer-Verlag, New York.
  • Platt, W.J., Hill, G.R. & Clark, S. (1974) Seed production in a prairie legume (Asragalus canadensis L.): interactions between pollination, predispersal seed predation, and plant density. Oecologia, 17, 5563.
  • Preston, C.D., Pearman, D.A. & Dines, T.D. (2002) New Atlas of the British and Irish flora. Oxford University Press, Oxford.
  • Preston, C.D., Telfer, M.G., Roy, D.B., Carey, P.D., Hill, M.O., Meek, W.R., Rothery, P., Smart, S.M., Smith, G.M., Walker, K.J. & Pearman, D.A. (2003) The Changing Distribution of the Flora of the United Kingdom: Technical Report. Centre for Ecology and Hydrology, Monks Wood, Cambridgeshire, UK.
  • Quinn, G.P. & Keough, M.J. (2002) Experimental Design and Data Analysis for Biologists. Cambridge University Press, Cambridge.
  • Quinn, R.M., Lawton, J.H., Eversham, B.C. & Wood, S.N. (1994) The biogeography of scarce vascular plants in Britain with respect to habitat preference, dispersal ability and reproductive biology. Biological Conservation, 70, 149157.
  • Rees, M. (1995) EC-PC comparative analyses? Journal of Ecology, 83, 891893.
  • Rheindt, F.E., Grafe, T.U. & Abouheif, E. (2004) Rapidly evolving traits and the comparative method: how important is testing for phylogenetic signal? Evolutionary Ecology Research, 6, 377396.
  • Schoenwald-Cox, C. & Buechner, M. (1991) Housing viable populations in protected habitats: the value of a coarse-grained geographical analysis of density patterns and available habitat. Species Conservation: A Population-Biology Approach (eds A. Seitz & V. Loeschcke), pp. 213226. Birkhäuser-Verlag, Basel.
  • Stevens, P.F. (2002) Angiosperm Phylogeny Website, Version 3. May 2002. http://www.mobot.org/MOBOT/research/APweb/[accessed 12 November 2002].
  • Stewart, A., Pearman, D.A. & Preston, C.D. (1994) Scarce Plants in Britain. JNCC, Peterborough.
  • Suarez, A.V., Holway, D.A. & Case, T.J. (2001) Patterns of spread in biological invasions dominated by long-distance jump dispersal: insights from Argentine ants. Proceedings of the National Academy of Sciences, 98, 10951100.
  • Telfer, M.G., Preston, C.D. & Rothery, P. (2002) A general method for measuring relative change in range size from biological atlas data. Biological Conservation, 107, 99109.
  • Thompson, K. & Hodgson, J.G. (1996) More on the biogeography of scarce vascular plants. Biological Conservation, 75, 299302.
  • Thompson, K., Rickard, L.C., Hodkinson, D.J. & Rees, M. (2002) Seed dispersal: the search for trade-offs. Dispersal Ecology (eds J.M. Bullock, R.E. Kenward & R.S. Hails), pp. 152172. Blackwell Science, Oxford.
  • Tilman, D. (1994) Competition and biodiversity in spatially structured habitats. Ecology, 75, 216.
  • Werner, P.A. & Platt, W.J. (1976) Ecological relationships of co-occurring goldenrods (Solidago: Compositae). American Naturalist, 110, 959971.
  • Westoby, M., Leishman, M.R. & Lord, J.M. (1995) On misinterpreting the ‘phylogenetic correction’. Journal of Ecology, 83, 531534.
  • Wigginton, M.J. (1999) British Red Data Books. 1. Vascular Plants, 3rd edn. JNCC, Peterborough.
  • Wilson, R.J., Thomas, C.D., Fox, R., Roy, D.B. & Kunin, W.E. (2004) Spatial patterns in species distributions reveal biodiversity change. Nature, 432, 393396.
  • Wright, I.J., Clifford, H.T., Kidson, R., Reed, M.L., Rice, B.L. & Westoby, M. (2000) A survey of seed and seedling characters in 1744 Australian dicotyledon species: cross-species trait correlations and correlated trait-shifts within evolutionary lineages. Biological Journal of the Linnean Society, 69, 521547.

Supporting Information

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

Appendix S1 Phylogenetic tree of rare and scarce British plants.

FilenameFormatSizeDescription
JEC_1123_sm_table.doc139KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.