Seed mass and the competition/colonization trade-off: a sowing experiment


*Correspondence: Mark Rees (


1  A seed-addition experiment using seven co-occurring annual plant species with a range of seed masses was carried out in a limestone grassland in South Wales.

2  If seedlings compete for establishment sites, then large seed size may confer enhanced competitive ability. However, the simple reciprocal relationship found between seed mass and per capita seed output showed that species producing larger seeds suffer reduced fecundity. Seed size may therefore act as a surrogate in a competition/colonization trade-off.

3  Equal numbers of seeds of all species were sown in a mixture over a range of densities. As sowing density increases, all species should reach a higher proportion of the available microsites. If large-seeded species are the best competitors they are expected to win all the sites they reach, and hence to occupy an increasing proportion of sites as sowing density increases.

4  The three species with the largest seeds made up 49% of individuals at low-density sown plots but 83% of individuals in high-density sown plots. In addition, seed mass and plant density were not correlated in unsown plots, but were strongly correlated in high-density sown plots. However, all small-seeded species maintained a presence in sown plots.

5  Although species were sown at random with respect to one another, individuals were up to five times more likely than expected to have a conspecific as a nearest neighbour. This could be caused by interspecific competition and/or by environmental heterogeneity that favours different species in different patches.

6  The results suggest that seedlings do compete for establishment sites and that large-seeded species generally win when in direct competition. In unsown areas small-seeded species win many sites by forfeit (because large-seeded species are strongly recruitment limited) but there may be a restricted subset of potential sites for which they are the best competitors and which they can win outright.


Nearly all papers on the evolution or ecological significance of seed size begin with two observations. First that, across the global flora, seed size varies over some 10 orders of magnitude, and secondly that, within species, seed size is remarkably constant (Harper et al. 1970; Haig 1989; Silvertown 1989). While the second observation implies that there is strong stabilizing selection on seed size, the first implies that the selected seed size varies widely between species.

An explanation for the apparent constancy within species was provided by the theoretical treatment of Smith & Fretwell (1974), who assumed that a plant has a fixed amount of resources to allocate to reproduction and that a decision must therefore be made concerning both the number and size of those offspring. Their model predicts that there will be a single optimum seed size that is evolutionarily stable (Lloyd 1987): individuals that produce seeds either smaller or greater than the optimum suffer reduced fitness, and consequently any observed variability in seed size is maladaptive (Lloyd 1987; Geritz 1995). However, it has been pointed out that the model fails to explain why the form of the seed size/offspring fitness function should vary so dramatically between species that share the same habitat (Geritz 1995; Rees & Westoby 1997).

However, if seedlings of other species form an essential part of the offspring's environment (for example through competition for establishment sites) and if seed size affects the competitive ability of seedlings (Pemadasa & Lovell 1974; Mack & Harper 1977; Law & Watkinson 1987), offspring fitness will depend not only on maternal provisioning of the species concerned, but also on the seed size of competing individuals (Geritz et al. 1988; Geritz 1995; Rees et al. 1996; Rees & Westoby 1997). Using these assumptions Rees & Westoby (1997) showed that, providing the advantage to larger seeds is capped in some way, this can lead to coalitions of species within which each species has a different seed mass. Because no single seed size is evolutionarily stable, new species can invade the community providing that their seed size is sufficiently different from that of resident species. If producing larger seeds results in reduced fecundity, the model is an example of a competition/colonization trade-off, in which seed mass determines both competitive and colonizing ability via the trade-off between seed size and seed number. Several theoretical treatments have demonstrated that competition/colonization trade-offs can allow the coexistence of a large number of species (Skellam 1951; Armstrong 1976; Hastings 1980; Tilman 1994).

It has been suggested that a competition/colonization trade-off based on seed mass might promote coexistence in sand-dune annual communities (Rees 1995; Rees et al. 1996); we explored this possibility with an overlapping group of species from a limestone grassland. The species have been observed to differ greatly in their seed masses (see Table 1), although it is not known whether large-seeded species do indeed suffer reduced fecundity. It is entirely possible that individuals of species that produce larger seeds will also capture more resources in total and hence produce more seeds as well as larger seeds, as is often found to be the case when such trade-offs are looked for within species (Venable 1992). In the natural community all species-pairs are spatially segregated (Turnbull 1998), meaning that all species experience an increase in the ratio of conspecific to heterospecific neighbours above that expected from their respective population sizes (Coomes et al. 1999). It has been suggested that such spatial segregation may result from local dispersal (Mahdi & Law 1987), although it may also indicate the presence of an underlying mosaic of patches with species specializing on different patch types (Mahdi & Law 1987; Law et al. 1993). In this case, seed size differences among species may represent such an adaptation to different types of microsite.

Table 1.  The seed masses of each of the sown species together with the estimated background seed production in the permanent quadrats (from Turnbull 1998). Background seed production was estimated from maps of all individuals in the permanent quadrats and their per capita seed outputs. Note that in 1995 (the year the experiment was carried out) background seed production of all species was very low
Background seed production (10 cm–2)
SpeciesSeed mass (g)19941995
Aphanes arvensis0.1594.770.003
Myosotis ramosissima0.1343.670.030
Veronica arvensis0.075948.011.80
Arenaria serpyllifolia0.054520.360.09
Cerastium diffusum0.0414145.501.27
Cerastium glomeratum0.034110.750.15
Saxifraga tridactylites0.0126120.810.491

A seed-sowing experiment was devised to assess the advantage to larger seeds, and to test specifically for a competitive hierarchy based on seed mass. Seed sowing experiments have been used before to demonstrate colonization limitation (Putwain et al. 1968; Gross & Werner 1982; Fowler 1986; Kelly 1989; Eriksson & Ehrlen 1992; Tilman 1997) but not to answer more detailed questions about community structure. In this experiment the effect of any trade-off between seed size and seed number is annulled by sowing mixtures of seed composed of equal numbers of all species at a range of densities. As sowing density increases, each species should reach an increasing proportion of the available microsites and so there is less potential for winning-by-forfeit (sensuHurtt & Pacala 1996).

If the community is structured by spatial niche differentiation, each species will have a competitive advantage in some microsites, and will therefore be expected to win some sites even when all species reach all sites. In this situation there should therefore be no reduction in the number of species present as sowing density increases (Fig. 1a). However, if there is a clear competitive dominant, it is predicted that, once its colonization limitation has been overcome, this species will occupy all microsites and exclude all others (Tilman 1994; Pacala & Rees 1998). Hence, as sowing density increases, any competitive dominant is expected to make up an increasing fraction of the total number of recruits (Fig. 1b). If greater seed mass confers enhanced competitive ability we expect the competitive hierarchy to be based on seed mass.

Figure 1.

Possible results from a hypothetical sowing experiment in a three-species community, involving the addition of low (L), medium (M) and high (H) density seed mixtures. In each case the seed mix is assumed to contain equal numbers of seeds of the three species. (a) Spatial niche differentiation: each species is the best competitor for some fraction of the available microsites. All species persist even at the highest seed input density when all species reach all sites. (b) Single competitive hierarchy: one species (solid bar) is the best competitor for all microsites. As sowing density increases this species reaches, and occupies, an increasing proportion of sites, eventually displacing all other species.

Materials and methods

Site and species selection

The site was on the Gower Peninsula, South Wales (National Grid Reference SS/423872). The area consists of Carboniferous limestone cliffs covered by shallow soils and the vegetation is classified within the National Vegetation Classification as CG1: Festuca ovinaCarlina vulgaris grassland (Rodwell 1992). These grasslands are noted for high densities of short-lived species, and a detailed study of the population dynamics of eight species in 30 60 × 40-cm permanent quadrats was initiated in 1994 and continued for 3 years. In these quadrats the locations of all individuals were recorded together with an estimate of each plant's fecundity. Six of these species were used in this experiment (Cerastium diffusum Pers., Arenaria serpyllifolia L., Saxifraga tridactylites L., Aphanes arvensis L., Veronica arvensis L. and Myosotis ramosissima Rochel) and a seventh species that was uncommon in all 3 years (Cerastium glomeratum Thuill.). Seed weights of all species are given in Table 1. All sown seed was collected locally during June 1995.

Seed size/number trade-off

A sample of 30 individuals of each species was taken from the area around the study site. Plants were selected subjectively to encompass a wide range of sizes. Plants were air-dried for several weeks and the seed pods dissected. For all plants with mature fruits, mean seed mass was estimated by weighing a sample of seeds from each individual. Where possible five seeds from each plant were weighed. Mean per capita seed production was estimated from the observed fecundity of individuals within permanent quadrats. This therefore represented the expected fecundity of an average individual.

In the simplest case, where all species have the same amount of resources (R) to allocate to reproduction, a trade-off between seed mass and seed number is expected:


which, on taking logs, gives:


If a trade-off exists in this form, regression of log seed number against log seed mass should yield a slope indistinguishable from –1.

Seed sowing

In March 1995, 10 60 × 40-cm quadrats were randomly positioned around the existing permanent quadrats. In each of the new quadrats all individuals of all sown species were recorded and removed prior to seed set. There was therefore no natural seed production by any sown species within the experimental quadrats during 1995. Only the central 20 × 20-cm square, divided into four 10 × 10-cm plots, was used for the experiment. One of the following treatments was assigned at random to each plot: (i) no seeds of any species; (ii) low seed density, 10 seeds of each species; (iii) medium seed density, 50 seeds of each species; (iv) high seed density, 200 seeds of each species.

All seeds were sown at the end of June 1995 (i.e. the time of natural seed fall) to ensure that any requirements for breaking dormancy were met. This also exposed the seeds, most of which germinate in the autumn, to predation through the summer. For all the sown species the highest sowing density exceeded the maximum recorded background seed rain (Table 1). The background seed rain varied over three orders of magnitude between species, and was negatively, although non-significantly, correlated with seed size in both the year prior to sowing (1994, rs = –0.643, NS, n = 7) and the year in which sowing took place (1995, rs = –0.536, NS, n = 7). This suggests that larger-seeded species are more strongly recruitment-limited. Note that Cerastium glomeratum has a relatively high background seed rain despite its low population density, because of its very high fecundity.

The plots were assessed in the year following sowing on two dates approximately 1 month apart. On the first date (20 April 1996) plants were beginning to fruit and the number of individuals of each species in each plot was recorded. On the second date (27 May 1996) plots were fully mapped and the fecundity of each individual was recorded. We analysed both the estimated number of plants establishing from sown seed and the effect of seed mass on final community composition. Both analyses were necessary because if, for example, smaller-seeded species have more substantial seed banks, they might show a poor response to sowing and yet still be numerically dominant in sown plots. The effect of sowing density on the proportion of sown seed establishing was also analysed to assess whether or not there was evidence of density-dependent emergence and/or establishment. The correlation between seed mass and plant density in the permanent quadrats was also assessed for comparison with the sown plots.

All parametric analyses were carried out using the glim 3.77 statistical package (Royal Statistical Society, London, UK). glim uses maximum likelihood techniques to estimate the parameter values and deviance as the generalized measure of goodness-of-fit (McCullagh & Nelder 1989). Most of the data were in the form of counts, and straightforward linear regression methods were therefore not appropriate (Crawley 1993). glim allowed us to specify a Poisson error distribution (which assumes that the variance is equal to the mean) and a log link that prevented the occurrence of negative fitted values. However, for most ecological data the variance is actually greater than the mean (over-dispersion) and the standard errors produced by a model assuming a Poisson error distribution are too small. In order to correct for this, we needed to rescale the model. Although such a scale parameter can be estimated empirically, for example using the ratio of residual deviance to degrees of freedom, it is recommended to use Pearson's χ2 (Aitkin et al. 1989). Where over-dispersion was severe (Pearson's χ2/residual degrees of freedom > 5 after model fitting and simplification), non-parametric tests were also carried out. Proportions were analysed using binomial errors and a logit link (Crawley 1993), and, where necessary, rescaling was carried out using the procedures outlined above.

Spatial patterns

Intraspecific aggregation

As the same number of seeds was sown into each quadrat, seed densities should have been equal, providing that the sown seed overwhelmed the seed bank. Intraspecific aggregation between sown quadrats could not be due to local dispersal effects and must primarily have been due to differences in environmental quality. For those species that responded strongly to the seed-addition treatment (where supplementary seed is likely to be the main source of propagules) the degree of aggregation between quadrats was quantified by calculating the index of dispersion:


where x is the mean number of plants per quadrat and s2 is the variance. If the counts are Poisson distributed, I(n – 1) has a chi-square distribution with n – 1 degrees of freedom (Elliott 1977). In order to assess the contribution of the seed bank to the spatial aggregation, the index of dispersion was calculated for both sown and unsown plots.


Species were sown randomly with respect to one another, therefore local dispersal could not be a cause of segregation in the resulting pattern of adults. Any segregation present must either have been due to underlying differences in patch quality or to competitive effects acting on germination or survivorship. We used a segregation index (Sij) based on the identity of the nearest neighbour:


where Nii is the number of individuals of species i with a nearest neighbour of type i, Nij is the number of individuals of species i with a nearest neighbour of type j, Ni is the population size of species i and Nj is the population size of species j (Dixon 1994). If species are well-mixed, values around unity are expected. Values in excess of unity indicate that the nearest neighbour is more likely to be a conspecific. Because of the small numbers of individuals involved, segregation indices could not be calculated for individual species-pairs. Instead, the extent of segregation between each species and all the remaining sown species was calculated across all sown quadrats. The significance of the observed value was determined using a Monte Carlo procedure. At each randomization, species’ names were assigned randomly to locations, and the segregation index recalculated. The number of values of Sij greater than or equal to the observed value was used to assign significance probabilities. Significance tests were based on 1000 randomizations. An edge correction was applied by discounting all individuals nearer to the plot edge than to another plant.


Seed size/number trade-off

Mean seed masses for individual plants are shown in Fig. 2. Although there was substantial variation within species, there was a clear separation between the species along the seed size axis. Across all species, average per capita seed production was related linearly to mean seed mass on a log scale (Fig. 3; F1,5 = 14.19, P < 0.05, n = 7). The slope of the line was not significantly different from –1 (t = 0.15, P > 0.05), which is consistent with a simple reciprocal relationship between seed number and seed mass. This implies that an average individual has roughly the same amount of resources to allocate to reproduction regardless of species identity, and that species producing larger seeds therefore suffer reduced fecundity.

Figure 2.

Phenotypic variation in seed mass. Each point is the mean seed mass of a single plant. The species are arranged in descending order of mean seed mass (log scale).

Figure 3.

Average per capita seed production (calculated over 2 years) plotted against mean seed mass (log scales). The fitted line is y = 1.096–0.961x (r2 = 0.74).

Responses to seed sowing

The total number of plants increased with sowing density on both recording dates, although there was considerable mortality between the two dates. However, examination of the counts from each plot revealed that there must also have been limited recruitment, particularly of Arenaria serpyllifolia. The increase in total plant density with sowing suggested that seeds from the seed bank were unable to fill all potential microsites, and that seed production in one year strongly influenced plant density in the following year.

In order to look at individual responses to the sowing treatments, the species were analysed separately. Inspection of the data showed that Cerastium diffusum and Arenaria serpyllifolia on at least one recording date achieved consistently higher densities in sown vs. unsown plots, but that the density achieved was independent of sowing density (Fig. 4). Model simplification was therefore attempted by combining the three seed-addition treatments into a single level, and refitting with the new two-level factor (sown and unsown; Table 2). No statistical analysis was carried out on either S. tridactylites or Cerastium glomeratum, as, although there were more plants in sown plots, the number of plants emerging in each case was low. Models for Aphanes arvensis, M. ramosissima and V. arvensis were shown to require four levels of sowing density at both sowing dates. For Cerastium diffusum, where there were significant effects of sowing at both recording dates, and for Arenaria serpyllifolia, where the effect was significant only at the second recording date, the model could be simplified to two levels of sowing (sown and unsown).

Figure 4.

The number of plants of each species in each sowing treatment at the first recording date (means and one standard error).

Table 2.  The change in scaled deviance associated with the removal of sowing density from the maximal model at the first and second recording dates (early and late). Density 4 is the value associated with four levels of sowing density; Density 2 is the value associated with two levels of sowing density (sown and unsown); Δ Density is the value associated with a simplification from four levels of sowing density to two
DateSourced.f.Aphanes arvensisMyosotis ramosissimaVeronica arvensisCerastium diffusumArenaria serpyllifolia
  • *

    P  < 0.05;

  • * *

    P  < 0.01;

  • * * *

    P  < 0.001.

EarlyDensity 43153.4***110.7***60.72***9.5*6.46
 Density 2162.74***38.95***13.58**9.48**1.112
  Δ Density290.67***71.74***47.14***0.020
LateDensity 43129.5***25.83***14.65**5.645.50
 Density 2150.95***12.19***0.395.53*5.25*
  Δ Density278.6***13.64**14.26**0.110.25

If large seeds have a higher probability of establishing, we expect a positive relationship between seed mass and the number of plants establishing from a given quantity of sown seed. To test this, the number of recruits of each species in each sowing treatment was pooled across all blocks and the number of plants emerging in the unsown plots subtracted. This gave an estimate of the number of plants establishing from sown seed. Seed mass was log transformed to improve linearity. The residuals still exhibited considerable over-dispersion, with scale parameters (estimated using Pearson's χ2) of 8.04 (early) and 5.6 (late). After correction for over-dispersion there were highly significant main effects of both density (early: F2,19 = 13.53, P < 0.001; late: F2,19 = 7.93, P < 0.005) and seed mass (early: F1,18 = 34.06, P < 0.001; late: F1,18 = 27.98, P < 0.001) on the number of recruits at both recording dates (Fig. 5). The interaction was not significant at either date (early: F2,17 = 1.84, P > 0.05; late: F2,17 = 2.94, P > 0.05). The significance levels should be treated as approximate given the high levels of over-dispersion. Because of this, the relationship between seed size and number of recruits at each density was also examined using Spearman rank correlation. There was a significant correlation between seed mass and number of recruits at the highest sowing density at each recording date but not at the lower sowing densities (early: low density, rs = 0.455, NS; medium density, rs = 0.493, NS; high density, rs = 1.000, P < 0.01; late: low density, rs = 0.127, NS; medium density, rs = 0.500, NS; high density, rs = 0.929, P < 0.01; n = 7 in all cases).

Figure 5.

The relationship between seed mass (log scale) and the estimated number of plants establishing from sown seed at the first recording date. Open circles: 10 seeds of each species; open squares: 50 seeds of each species; closed circles: 200 seeds of each species. The fitted lines are y10 = exp(4.855 + 1.27x); y50 = exp(5.876 + 1.27x); y200 = exp(6.896 + 1.27x); r2 = 0.81.

Community composition

If small-seeded species have more substantial seed banks, they might be expected to respond less strongly to the sowing treatments. However, they could still be numerically dominant in sown plots. The community composition in each sowing treatment is shown in Fig. 6. There was no significant correlation between seed mass and plant density in the unsown plots (rs = 0.400, NS, n = 7) or in the permanent quadrats in 1994 (rs = 0.107, NS, n = 7) and 1995 (rs = 0.072, NS, n = 7). However, seed mass and plant density appeared to be correlated in sown plots, except at low density (Fig. 6), with a strong and significant correlation at the highest sowing rate (low density: rs = 0.500, NS; medium density: rs = 0.750, NS; high density: rs = 0.964, P < 0.01; n = 7 in all cases). Across sowing treatments there was no tendency for the absolute number of recruits of the smaller-seeded species to decrease with increased sowing density (with the exception of the smallest, S. tridactylites, which was almost absent from the high-density treatment). Large-seeded species were therefore unable to exclude small-seeded species at these sowing rates. As sowing density increased, large-seeded species comprised an increasing fraction of the community (Fig. 6). For example, the three species with the largest seeds constituted 83% of individuals at high sowing density but only 49% of individuals at low sowing density. They constituted 39% of individuals in the permanent quadrats.

Figure 6.

Community composition in each of four sowing treatments. Species are arranged in decreasing order of seed size, with Aphanes arvensis at the far left and Saxifraga tridactylites at the far right. There was little correlation between seed mass and proportion of individuals in the zero seed treatment, but a strong correlation at the highest sowing rate (see text).

Density dependence


For each of the species where analysis of the sowing treatments was possible, the number of plants emerging from sown seed in each plot was estimated by subtracting the number of plants in the unsown plot from the number of plants in each of the sown plots. Occasionally, the number of plants in the unsown plot exceeded the number of plants in sown plots and in this case no plants were considered to have emerged from sown seed. Of the five species analysed, there was evidence of negative density-dependent emergence and/or survivorship on at least one of the recording dates for Cerastium diffusum (early: F2,20 = 11.6, P < 0.01; late: F2,20 = 10.53, P < 0.01), Arenaria serpyllifolia (late: F2,20 = 3.83, P < 0.05) and M. ramosissima (early: F2,20 = 4.79, P < 0.05).


There were few plants remaining by the second recording date, hence estimates of average fecundity in each plot were based on a small sample of individuals. Average fecundity was analysed using normal errors and weighted by the number of individuals on which the mean was based. There were no significant effects of sowing density for any of the species analysed.

Population growth rates

Seed production from sown seed was estimated by subtracting the number of seeds produced in unsown plots from the number of seeds produced in sown plots. The estimated population growth rate was then calculated as the quotient:


The mean value of λ declined with sowing density for all species except Cerastium glomeratum (Fig. 7). Only three of the seven species showed population expansion (λ ≥  1) in low-density plots. However, variability was particularly high at low density, and for all the common species there were blocks where the growth rate was considerably in excess of unity. The remaining species, Cerastium glomeratum, was rare at the site, and high seed input densities seemed to be required for its persistence. It may not therefore constitute a self-sustaining population. The mean value of λ at low density appeared to be higher for species with low seed mass, while the mean value of λ at high sowing density appeared to be higher for species with high seed mass. Because of the extreme over-dispersion, the relationship between seed mass and λ was analysed using Spearman rank correlation. Correlation coefficients were calculated excluding Cerastium glomeratum, as the population of this species may not be self-sustaining. Although none of the correlation coefficients was significant, the sign of the correlation coefficient was negative at low density and positive at high density (low density: rs = –0.693, NS; medium density: rs = –0.495, NS; high density: rs = 0.573, NS).

Figure 7.

Estimated population growth rate of seven species at three sowing densities: 10, 50 and 200 seeds of all species. Means were calculated from 10 quadrats, and error bars are one standard error (+ indicates extreme values).

Spatial patterns

All species whose response to the sowing treatments was analysed showed significant aggregation between quadrats at both sowing dates in sown plots (Table 3). Cerastium diffusum and V. arvensis were also significantly aggregated between unsown plots, although with a lower index of dispersion. The segregation indices ranged from 1.26 to 5.45, indicating that the nearest neighbour was up to five times more likely to be conspecific than any of the other sown species (Table 4). The segregation index was significantly in excess of unity for the three species that responded most strongly to the seed additions. Thus, although species were sown randomly with respect to one another, individuals were more likely to have a conspecific nearest-neighbour in the resulting pattern of adults.

Table 3.  Intensity of aggregation between quadrats in both sown and unsown plots at two recording dates (early and late). The index of dispersion (I) was calculated for all species in which there were significantly more plants in sown plots
  • *

    P  < 0.05,

  • * *

    P  < 0.01,

  • * * *

    P  < 0.001.

Aphanes arvensis1019.819***1.60315.872***1.222
Myosotis ramosissima107.264***1.0003.624***1.778
Veronica arvensis105.617**2.231*2.823**2.827**
Arenaria serpyllifolia10 –   –   3.259***0.889
Cerastium diffusum109.746***3.296***6.242***3.111***
Table 4.  The degree of segregation between each species that responded to the sowing treatments and the remaining sown species. The segregation index was based on the identity of the nearest neighbour. Significance levels were assigned by conducting 1000 rounds of random labelling. The significance level was the proportion of randomizations that yielded values of the segregation index greater than or equal to that observed
SpeciesSegregation indexSignificance level
Aphanes arvensis3.16  < 0.001
Myosotis ramosissima4.42  < 0.001
Veronica arvensis5.45  < 0.001
Arenaria serpyllifolia2.51  > 0.11
Cerastium diffusum1.26  > 0.25


Seed size variation

In common with other studies we found considerable phenotypic variation in seed size within species (Michaels et al. 1988; Stocklin & Favre 1994). However, this variation never entirely overlapped that of adjacent species in the seed size hierarchy. In addition, as fully ripe seeds are difficult to identify unequivocally, the extent of overlap between species may have been over-estimated due to the inevitable inclusion of unripe seeds in weighed samples. The partial separation of species along a seed size axis suggests that each species has found a different solution to the seed size problem.

Among the seven species used in the experiment, we found a simple reciprocal relationship between mean seed mass and average per capita seed production. Small-seeded species therefore have a fecundity advantage among this group of specie, where there is a clear cost to producing larger seeds (Shipley & Dion (1992). Such a trade-off occurs because all species appear to have roughly the same total reproductive biomass. Such a clear trade-off between seed size and seed number is less likely to occur when reproductive biomass varies greatly between species. However, other allocation trade-offs, such as root/shoot ratios, have been linked to both competitive and colonizing abilities (Gleeson & Tilman 1990; Tilman 1994).

If species compete for establishment sites, large-seeded species must have some advantage to counterbalance their reduced seed production. This advantage is clearly an increased probability that each seed will germinate and establish successfully: the estimated proportion of sown seeds that established successfully was strongly correlated with seed mass at the highest sowing rate. However, this advantage could take a number of forms. First, small-seeded species could suffer higher levels of predation. It is known that seed predation is often size-dependent (Blate et al. 1998), but it is difficult to imagine that seeds weighing only 0.01 g would suffer the highest rate of attack. Secondly, there could be a strict competition/colonization trade-off, with the largest species being the best competitor for all sites. Thirdly, there could be some kind of microsite specialization, with safe sites for the large-seeded species being much more frequent than those for small-seeded species. This possibility is in agreement with numerous experimental studies that have shown that larger-seeded species are better able to establish and survive under a wide range of conditions (reviewed in Westoby et al. 1997).

Community structure

In a strict competition/colonization trade-off, the same range of microsites is potentially available to all species (Fig. 8a) and the competitive hierarchy is the same across all sites (Hastings 1980; Tilman et al. 1994). If the trade-off is based on seed mass the species producing the largest seeds is expected to be the best competitor. Competition occurs between seedlings within microsites and the seedling from the largest seed is predicted to claim the site. Although small-seeded species lose out in direct competition, their fecundity advantage ensures that they are present in a higher fraction of sites and establish in those from which bigger seeds are absent. Coexistence depends critically on inescapable allocation trade-offs that prevent species from being both good competitors and good colonizers.

Figure 8.

Two extreme theoretical views of microsite use in a seven-species guild where species 1 has the largest seeds and species 7 the smallest. (a) Competition/colonization trade-off. (b) Extreme spatial niche differentiation. In (a) all species have the same microsites potentially available to them. In (b) each species specializes on an exclusive range of the available variability, but the biggest-seeded species (species 1) has the largest range of available microsites.

The sowing experiment presented here supports this hypothesis in several respects. Larger-seeded species not only established better from sown seed but also became numerically dominant at high sowing rates. The three species with the largest seeds made up 83% of the total in high-density plots, and community composition became more strongly correlated with seed mass as sowing density increased. However, the competition/colonization trade-off requires that the best competitor is the most strongly recruitment limited at the population level. Because all natural seed inputs were prevented, this experiment is not a direct test of colonization limitation. However, mean natural seed production in a plot of similar area was negatively (albeit non-significantly) correlated with seed size. In addition, the highest sowing rate was 50 times higher than the mean natural seed production for the two species with the largest seeds. Projected population growth rates of large-seeded species (the quotient seeds produced/seeds sown) were always less than unity in high-density plots, indicating that the densities achieved at high sowing rates cannot be maintained. It therefore appears that the large-seeded species are indeed strongly recruitment limited at the population level.

If a strict competition/colonization trade-off operates we expect the best competitor to exclude all other species at sufficiently high sowing rates. We did not observe exclusion of any species (although only one S. tridactylites plant established from a total of 2000 seeds sown in high-density plots). It also seems likely that higher sowing rates than those used here would have led to increased establishment of larger-seeded species as there was little evidence of density-dependent emergence and/or survivorship for these species. However, given that small-seeded species were not excluded, we cannot discount the possibility that seed size variation is a result of microsite specialization.

Under an extreme niche-specialization hypothesis each species has an exclusive range of microsites available (Fig. 8b); however, to account for the correlation between seed mass and plant density at high sowing rates, the frequency of safe sites must be assumed to increase with seed mass. Spatial segregation would arise if adjacent microsites were more similar to each other than distant microsites and hence more likely to be occupied by individuals of the same species. If small-seeded species specialize on a limited range of microsites, high fecundity may be favoured, in order to increase the probability of successful dispersal to relatively scattered safe sites, and this may select strongly against any increase in seed size. However, in sand-dune annual communities small-seeded species are numerically dominant (Rees 1995), and here Cerastium diffusum (one of the species with the smallest seeds) was the most abundant species in the permanent quadrats. It seems unlikely that small-seeded species could achieve such high population densities if the range of microsites available to them was so limited.

Alternatively, there may be some combination of competition/colonization trade-off and microsite specialization with small-seeded species establishing in a much broader range of sites in the absence of large-seeded competitors (Fig. 9). Under this scenario winning-by-forfeit is common at low seed input densities and small-seeded species are able to exploit a wide range of sites. However, at high seed input densities, the scope for winning-by-forfeit is vastly reduced, and small-seeded species are constrained by large-seeded competitors. Different types of microsites may either form a fixed template or vary from year to year. When microsite quality varies between years the conditions for coexistence become more stringent, as species need to colonize distant sites that have become favourable, and more seeds are lost by dispersing into now unfavourable sites (Comins & Noble 1985; Hurtt & Pacala 1996).

Figure 9.

Hypothetical microsite use by seven species ranked in order of decreasing seed mass. In this case each species has a wide range of sites that it can occupy in the absence of other species (dotted line) and a smaller range of sites that it can occupy in the presence of competitors (solid line). Smaller-seeded species can only win sites on the dotted line if larger-seeded species are absent; however, sites on the solid line are won outright.

A clear way to distinguish between these alternatives is to include single-species sowings in an experiment of this kind. We could then see whether small-seeded species achieved higher densities in single-species plots than in mixtures, providing more compelling evidence of competitive exclusion. However, the comparison with the background community (where there is no correlation between density and seed mass) demonstrates that observed community structure is strongly affected by colonization limitation.

Spatial structure

The dominant spatial patterns observed in permanent quadrats (intraspecific aggregation and segregation; Turnbull 1998) were also observed within sown plots. It therefore seems unlikely that local dispersal is solely responsible for pattern generation in this community (Mahdi & Law 1987). However, local dispersal may affect population persistence by ensuring that seeds land in favourable microsites. When seeds were dispersed randomly (during the sowing experiment) only three of the seven species showed population expansion from low density. However, all species (except Cerastium glomeratum) increased from low density in at least one sown quadrat. Almost all species therefore have the potential to increase when rare in part of the habitat, and local dispersal may increase the probability that seeds remain within favourable patches. Cerastium glomeratum, the only species that did not achieve positive population growth in any quadrat, was unusual in that it showed higher population growth rates in plots with higher levels of seed addition. The population may therefore not be self-sustaining at this site, perhaps because of the extreme scarcity of suitable microsites, and may instead depend on immigrant seed from populations in surrounding field margins. Cerastium glomeratum is the only one of the sown species with such a potential source of immigrant seed. Sink populations of annual plants maintained by immigration have been reported before (Keddy 1982; Watkinson 1985).

Concluding remarks

In conclusion, large-seeded species have an establishment advantage over small-seeded species, but they also produce fewer seeds. Large-seeded species greatly increased their abundance relative to small-seeded species once their colonization rate had been experimentally increased above mean background levels. Although small-seeded species were greatly suppressed they were not entirely eliminated from high-density sown plots. To account for the relative abundance of small-seeded species in natural communities, we suggest they probably win many sites by forfeit, although there may be a small fraction of microsites for which they are the best competitors. The spatial patterns in sown plots were remarkably similar to those in permanent quadrats, indicating that local dispersal is not the primary determinant of observed patterns (as has been suggested for perennial species in a limestone grassland; Mahdi & Law 1987). However, local dispersal may ensure that seeds land in favourable microsites.


This work was carried out while L. A. Turnbull was in receipt of a studentship from the Natural Environment Research Council. We thank The National Trust for permission to work at the site. Liz Manley helped enormously with preparation of the manuscript, and David Coomes and two anonymous referees gave valuable comments. Finally, we are grateful to Lindsay Haddon for greatly improving the flow of the manuscript, although we should emphasize that any outstanding errors are ours.

Received 13 May 1998revision accepted 27 April 1999