Herbivores, seed banks and seedling recruitment in mesic grassland

Authors

  • G. R. Edwards,

    Corresponding author
      *Present address and address for correspondence: AgResearch Grasslands, Tennent Drive, Private Bag 11008, Palmerston North, New Zealand (fax 0064 6351 8032; e-mail edwardsg@agresearch.cri.nz).
    Search for more papers by this author
  • M. J. Crawley

    1. Department of Biology, Imperial College at Silwood Park, Ascot, Berks SL5 7PY, UK
    Search for more papers by this author

*Present address and address for correspondence: AgResearch Grasslands, Tennent Drive, Private Bag 11008, Palmerston North, New Zealand (fax 0064 6351 8032; e-mail edwardsg@agresearch.cri.nz).

Abstract

1 An experiment was carried out in a species-poor acid grassland to determine the effect of insect, mollusc and rabbit herbivory on the size and composition of the seed bank and on seedling recruitment from the seed bank and seed rain. From 1991 to 1997, insects and molluscs were excluded with pesticides, and rabbits with fences. Seedling recruitment was monitored over 22 months in gaps established in the vegetation in summer 1995.

2 The most common species recorded from the seed bank in early summer 1995 were dicots (17 species), but perennial grasses (five species) were numerically the most abundant (65% of total). There was no relationship between the species composition of the seed bank and the established vegetation.

3 The size of the seed bank of eight species was greater on fenced plots, a result that reflected increased seed rain where rabbits were excluded. Insects and molluscs had no effect on the size of the seed bank of any species. The number of species in the seed bank was not affected by any of the herbivore exclusions.

4 A comparison of seedling emergence in gaps formed over the original soil with gaps where the soil had been sterilized indicated that only Galium saxatile and Cytisus scoparius recruited from the seed bank. Seedling recruitment was almost entirely derived from the recent seed rain, was dominated by the most abundant perennial grasses in the vegetation (Festuca rubra and Holcus lanatus), and had a species composition that resembled the established vegetation. Results highlight that the potential for seedling establishment in gaps to bring about vegetation change in this grassland is low.

5 Six species had higher seedling densities on rabbit-fenced plots, but the significant effect of fencing disappeared by plant maturity for most species. Survival of seedlings was lower on fenced plots where non-grazed biomass accumulated, so that after 22 months Agrostis capillaris was the only species with more plants present where rabbits were excluded. Rumex acetosa and Stellaria graminea showed higher seedling emergence where molluscs were excluded. More seedlings of Rumex acetosa were also found where insects were excluded. These invertebrate effects were still evident at plant maturity.

Introduction

There is a growing view regarding grasslands that small-scale disturbances that disrupt the dominant perennial cover (e.g. animal diggings, death of perennial ramets) act as important microsites (‘vegetation gaps’) for seedling recruitment (Jalloq 1975; Rapp & Rabinowitz 1985; Milton et al. 1997) and may be central to the maintenance of plant species richness (Grubb 1977; Lavorel et al. 1994). Within these microsites, seedling recruitment may occur from seed derived from the persistent seed bank (i.e. seeds deposited or produced at the site more than a year earlier; Thompson & Grime 1979) or from seed derived from the current or previous year (i.e. seed rain or transient seed bank). Most grasslands have a large persistent seed bank, often with a species composition that does not resemble the above-ground vegetation (Thompson & Grime 1979), and it is well documented that this seed can dictate the successional trends that occur following large-scale disturbances (e.g. cultivation; Crawley 1990a). However, for relatively undisturbed grasslands, where recruitment microsites are fewer and smaller, the role played by seedling recruitment from the seed bank in the persistence of species and in vegetation change, and how important it is compared to the recent seed rain, is less clear (Sarukhan 1974; Bullock et al. 1994; Pakeman et al. 1998). The extent of seedling recruitment from the seed bank and seed rain not only has significance to our general understanding of how plant species richness is maintained, but also has applied significance for attempts to restore species richness in species-poor grasslands (Wells et al. 1989). For instance, is it possible for changes in species composition to occur via recruitment from a seed bank that differs in composition from the current vegetation, or do species novel to the vegetation have to be sown?

The importance of herbivores in determining the abundance of plant species is a central issue in ecology (Crawley 1997) and seedling recruitment into gaps in the perennial vegetation is one area where herbivores could, in principle, play an important role (Hanley et al. 1996). Changes in plant recruitment in gaps might result from herbivores altering the number or species composition of seeds in the seed rain and seed bank (e.g. by selective flower or seed predation; Crawley 1983; Louda 1989; O’Connor & Pickett 1992; Louda & Potvin 1995), from direct, selective grazing of seedlings (Crawley 1990b), or from competitor release of seedlings following the defoliation of neighbouring plants (Hanley et al. 1995).

Grasslands are habitats for a wide variety of herbivores including insects, molluscs and vertebrates. As these different herbivores are likely to vary in their preference for particular plant species, in the plant part or life stage they attack (e.g. seeds, seedlings, mature plants), in their spatial and temporal pattern of defoliation, as well as in the impact of their defoliation (Crawley 1983, 1997), we would expect their effects on the seed bank and seedling recruitment to vary. There are several well-documented cases of insects (Clements & Henderson 1979; Brown 1985; Brown & Gange 1989; Louda & Potvin 1995) and molluscs (Hanley et al. 1995, 1996) affecting recruitment in natural plant populations, but these are greatly outnumbered by studies showing the profound impact of vertebrates (references in Crawley 1989, 1997). There are few studies that have documented the relative impact of vertebrate and invertebrate herbivores on seed banks and seedling recruitment within the same plant community (Van Leeuwen 1983; Gibson et al. 1987; Hulme 1994a; Palmisano & Fox 1997), and this represents a major gap in our understanding of the role of herbivores, seed banks and seedling recruitment in shaping plant community structure.

Here we describe work on seed banks, herbivory and seedling recruitment in vegetation gaps in a relatively undisturbed, mesic grassland. The work set out to address three related questions.

1  Does any seedling recruitment into gaps occur from seed derived from the persistent seed bank, and what is the relative importance of this compared to seedling recruitment from seed derived from the recent seed rain?

2  What is the relative importance of insect, mollusc and rabbit herbivory in determining the size and composition of the seed bank?

3  What is the relative importance of insect, mollusc and rabbit herbivory in determining the density and survival of seedlings that emerge?

Materials and methods

Experimental site

The study was carried out in a long-term field experiment set up in June 1991 in Nash’s Field, Silwood Park, Berkshire, UK (National Grid reference 41/944691). Nash’s Field is a 6-ha species-poor grassland on acid, sandy soil, with a long history of rabbit (Oryctolagus cuniculus L.) grazing (National Vegetation Classification MG6; Rodwell 1992). Species have been lost progressively from this grassland as a result of several decades of intense rabbit grazing, following the recovery of rabbit numbers after myxomatosis in the 1950s (M.J. Crawley, unpublished data). A combination of competitive exclusion by grazing-tolerant and unpalatable plants (e.g. Luzula campestris, Galium saxatile and Rumex acetosella), reduced seed production by species that are both palatable and seed-limited (Centaurea nigra and Arrhenatherum elatius;Edwards & Crawley 1999), and direct, selective grazing of some species (e.g. Trifolium spp., Lotus corniculatus and Crepis capillaris) probably accounts for why certain species are missing from parts of the grassland (Crawley 1990a). Nash’s Field is surrounded by oak (Quercus robur) and birch (Betula pendula) woodlands and a bracken (Pteridium aquilinum) stand, which all act as harbourage for rabbits. Visual inspection of the grassland before treatments began showed that the dominant grass species were Agrostis capillaris, Festuca rubra and Holcus lanatus, and the dominant herb species were Galium saxatile and Rumex acetosella. Soil disturbance at the site is caused primarily by European moles (Talpa europaea L.) and rabbits. Molehills and rabbit scrapes covered 2–3% of the soil surface each year and formed locally important, competition-free seed beds (G.R. Edwards, unpublished data). Nash’s Field experiences an average annual rainfall of 653 mm with little seasonal pattern. There was drought in the spring–summer period of the 2 years when the experiment was conducted (1995 and 1996; Edwards & Crawley 1999). Additional information on Nash’s Field may be found in Crawley (1990a). Nomenclature follows Stace (1997).

Experimental design and treatments

A six-factor factorial experiment was laid out in a split-plot design. Experimental treatments were insect herbivory (with and without insecticides), mollusc herbivory (with and without molluscicide), rabbit grazing (with and without rabbit fences), soil pH (with and without lime), plant competition (grasses or herbs removed with selective herbicides) and soil fertility (with and without N, P, K and Mg fertilizers). The study reported here examined the effects of insect, mollusc and rabbit herbivory on the size and composition of the seed bank and seedling recruitment; no measurements were made of the seed bank or seedling recruitment for the lime, plant competition and fertilizer treatments. Only the insect, mollusc and rabbit exclusion treatments are described in detail in this paper; brief details of other treatments are given for clarity.

The experiment was laid out in two blocks, one that was judged to be in a dry area of the field and one that was judged to be in a slightly more moist area. In each block, four whole plots, each measuring 22 m × 44 m, were laid out and were randomly allocated to one of the following treatments: (i) insecticide applied; (ii) molluscicide applied; (iii) insecticide and molluscicide applied; and (iv) no insecticide or molluscicide applied. Insects were suppressed by spraying the above-ground vegetation with dimethoate (dimethoate-40, Atlas Interlates, Glasshouse Farm, Bradford, BD12 0JZ, UK) at 336 g active ingredient ha−1 and chlorypyrifos (Spannit, PBI, Britannica House, Herts, EN8 7DY, UK) at 240 g active ingredient ha−1. Both are organophosphorus insecticides, dimethoate being contact and systemic, chlorypyrifos being contact and ingested. Molluscs were suppressed by applying pellets of metaldehyde (Mifaslug, FCC, Thorn Farm, Worcs, WR7 4LJ, UK) at 960 g active ingredient ha−1. Metaldehyde is a molluscicide bait for controlling slugs and snails. In each year from 1991 to 1996, all pesticides were applied three times during the spring–summer period in April, May and June and once during autumn in October.

We conducted separate glasshouse experiments for many of the species in Nash’s Field to ensure that the pesticides had no undue positive or negative effects on seedling emergence and seedling growth. In summary, we found that the application of the above insecticides, at the rates and frequencies used in the study, increased seedling emergence of five species (Agrostis capillaris, Arrhenatherum elatius, Centaurea nigra, Lotus corniculatus and Rumex acetosa) and increased seedling growth of one species (Trifolium repens). Application of the above molluscicide, at the rates and frequencies used in the study, decreased seedling emergence of one species (Agrostis capillaris) and reduced seedling growth of one species (Plantago lanceolata). In light of these findings, we interpret results of the effect of invertebrate herbivory on seedling recruitment with caution.

One half (plot size 22 m × 22 m) of each invertebrate herbivory whole plot was randomly allocated to be fenced to exclude rabbits and larger vertebrates in June 1991, giving eight grazed and eight fenced plots. The rabbit fences were 1 m high and were constructed of 3-cm square wire-mesh supported by 10-cm diameter posts every 4 m. The wire mesh was buried 5 cm deep, with the bottom 15 cm of wire turned outwards toward the rabbits so that they encountered the wire as soon as they started digging (Crawley 1990a). The fences were highly effective in excluding rabbits but larger vertebrates such as muntjuc (Muntiacus reevesi Ogilby.) and roe (Capreolus capreolus L.) deer could easily jump the fences. Furthermore, the fences did not exclude moles, which tunnelled under the fences, or rodents (e.g. seed predators like woodmouse (Apodemus sylvaticus L.), which entered through the mesh (Hulme 1994b). The fenced plots were managed as a hay meadow, with a single hay cut being taken from the plots using a hand-held sickle bar mower in early August each year. By harvest time most of the species present had finished flowering and dispersed their seed (M.S. Heard, unpublished data). All of the cut herbage was raked and removed from the plots to prevent any accumulation of dead organic matter.

Within each fenced and grazed plot, two plots, each measuring 18 m × 8 m, were laid out, and one plot was randomly allocated to be limed. There was a 2-m guard strip around the outside of each plot and a 2-m gap between the limed and unlimed plots. The limed and unlimed plots were further divided into plots where plant competition and fertilizer treatments were applied.

Vegetation survey

The cover of each species was determined in early June (summer) 1995 by 300 point quadrats taken at 5-cm intervals along a 15-m tape placed in the 2-m wide dividing area between limed and unlimed plots of each rabbit-grazed and rabbit-fenced plot. This area received the herbivory treatments, but was not limed, treated with herbicide or fertilized. The pin was lowered through the vegetation from the top of the canopy to the ground surface and contacts of the pin with each species were recorded. If a particular species was touched twice, only the first hit was recorded.

Flowerhead survey

Flowerheads of each species were counted in late June 1995 in four 0.5 m × 0.5 m quadrats placed at random in the 2-m wide dividing area between limed and unlimed plots. In some cases flowers were counted, whereas in others flowering stems were counted (see Table 2 for the details for each species).

Table 2.  The main effects of exclusion of insects, molluscs and rabbits on flowerhead production (per m2) in June 1995. Arithmetic means without standard errors are presented as all tests of significance were done on a log-transformed scale. The significance of the difference between the means is shown in the column to the right of each pair of means. (*P<0.01, NS=not significant. Other species did not occur frequently enough to be analysed)
 InsecticideMolluscicideRabbit fencing
SpeciesNoneSprayed NonePellets GrazedFenced 
  • †  

    Counts based on number of flowering stems (inflorescences).

  • ‡  

    Counts based on number of flowers.

Only present in the vegetation
Achillea millefolium1.01.5 0.52.0 1.01.5 
Arrhenatherum elatius4.22.6 3.13.6 2.64.2 
Crepis capillaris0.02.4 1.01.4 0.51.9 
Festuca rubra16.814.9NS13.517.8NS4.826.5 *
Hieracium pilosella0.52.1 1.01.6 0.52.1 
Hypochaeris radicata0.50.5 0.01.0 0.01.0 
Present in vegetation and seed bank
Agrostis capillaris18.217.7NS18.717.2NS12.224.3 *
Anthoxanthum odoratum8.35.7NS7.36.7NS3.110.9 *
Galium saxatile11.49.0NS10.010.5NS19.04.8 *
Holcus lanatus12.515.4NS11.616.4NS6.721.2 *
Holcus mollis0.52.1 1.01.6 0.52.1 
Leontodon autumnalis15.011.9NS16.810.1NS4.322.6 *
Lotus corniculatus1.03.1 1.62.6 0.53.6 
Luzula campestris11.99.5NS8.612.8NS11.99.5NS
Plantago lanceolata4.87.2 5.36.7 2.49.6 
Ranunculus repens1.00.5 0.51.0 0.01.6 
Rumex acetosa7.39.3NS7.88.8NS2.614.0 *
Rumex acetosella4.35.2 3.85.7 6.73.3 
Senecio jacobaea1.01.6 0.52.1 2.10.5 
Stellaria graminea15.618.2NS15.018.7NS9.923.9 *
Trifolium repens1.01.0 0.51.5 0.51.5 
Veronica chamaedrys3.14.2 4.23.1 2.15.2 
Vicia sativa4.46.7 4.97.1 2.29.3 

Seed bank samples

We define the seed bank as seeds, at or beneath the soil surface, that are capable of germination (Sagar & Mortimer 1976). Soil samples were collected in late spring (May) 1995, 4 years after the herbivore exclusion treatments began. We chose to collect the samples in May so that two main periods of seedling germination (autumn 1994 and spring 1995) had passed since the last seed dispersal, and to be prior to any fresh input of seed. Thus, at sampling the seeds remaining in the soil were approaching at least 1 year of age, and so were a good estimate of the between-year (persistent) seed bank (Thompson & Grime 1979). We did not attempt to estimate the within-year (transient) seed bank by taking soil samples throughout the year (Thompson & Grime 1979).

Seed bank samples were collected along an 18-m transect located up the centre of the 2-m wide dividing area between limed and unlimed plots. Starting 1 m along the transect, 16 soil cores, each 5 cm in diameter, were taken to a depth of 16 cm at approximately 1.1-m intervals. The total area sampled per rabbit-fenced or rabbit-grazed plot (0.031 m2) exceeded the minimum area per plot (0.02 m2) that Forcella (1984) found necessary for adequate assessment of seed banks. On removal from the soil, the core was carefully divided into a 0–8-cm and a 8–16-cm section, and samples for each plot were then bulked.

The bulked samples were broken down by hand into a fine crumble, and roots, rhizomes and any stones were removed. The samples were spread evenly in a 0.5–1-cm layer over 3–4 cm of sterilized soil (John Innes no. 1, sterile seed mix) in 40 cm × 30 cm × 5 cm plastic seed trays. The trays were placed randomly on benches in a glasshouse in late May (late spring) 1995 and their position re-randomized at 2-week intervals. The trays were maintained under glasshouse conditions of natural light and temperatures and were watered every 2–3 days. It is noteworthy that high temperatures occurred in the glasshouse (> 30 °C) in summer, which may have inhibited germination of some species. Six control trays of sterile soil were interspersed throughout the glasshouse to detect potential seed bank contaminants, either from the sterilized soil or immigrants. One Salix cineria seedling was detected in these trays, but no such seedlings emerged in the soil seed bank samples. All seedlings (germinable seeds) that emerged were identified, counted and removed every week for the first 2 months and then every 3 weeks after that; observation of seedlings during watering indicated that very few seedlings (< 0.5%) died between seedling counts. Any plants that could not be identified at the seedling stage were grown on until identification was possible. The soils were stirred every 4 months to expose ungerminated seeds following the cessation of the initial flush of germinants. Monitoring continued for 20 months in total to ensure that all dormant seeds had the opportunity to germinate.

Seedling recruitment from the seed bank and seed rain

Along the same transect from which soil samples had been taken, eight sampling sites were marked out at approximately 2-m intervals in mid-May (1995). Four sites were allocated at random to a ‘seed bank + seed rain’ treatment and four to a ‘seed rain’ treatment (Bullock et al. 1994). For the ‘seed rain + seed bank’ treatment, the above-ground vegetation, including buried shoot bases, was removed in a 15 cm × 15-cm square. Roots were severed to a depth of 10 cm around the outside of the square. For the seed rain treatment, a 15 cm × 15 cm square of soil was removed to a depth of 10 cm, taken back to the laboratory where the vegetation was removed, and sterilized by methyl bromide fumigation. The soil was then returned to the site from which it had been taken. Methyl bromide fumigation has been shown to be highly effective in killing buried seeds, shoots and roots, but it also kills soil invertebrates and fungi (M.J. Crawley, unpublished data). Seedling recruitment from the persistent seed bank was measured by comparing recruitment in the ‘seed rain + seed bank’ treatment with recruitment in the seed rain treatment. The gap size dimensions used were roughly equivalent to the size of the gap created by molehills and rabbit scrapes in the grassland (G.R. Edwards, unpublished data).

Our definition of successful seedling recruitment is that a seedling must survive to at least 1 year of age or to flowering (if this occurs within 12 months of germination). Thus, emergent seedlings do not necessarily become recruits, and recruitment can be considered to be comprised of two stages: (i) germination and emergence, followed by (ii) seedling survival and growth (Cook 1980). To investigate both these processes, we followed the emergence and subsequent fate of seedlings in the gaps until February 1997, 22 months after the creation of the gaps. Seedling censuses were conducted in August and November 1995, February, May, August and November 1996, and February 1997. At each census, any new seedlings in a gap were tagged with a tooth pick inserted into the soil nearby and their location was marked on a map of the quadrat. To help in the tagging of grass seedlings, loops of wire were placed around the plants when they reached 6 months of age. At each subsequent census, seedlings were recorded as either live or dead, with all missing seedlings recorded as dead. It was not possible to assign a cause of death to seedlings as most seedlings were either missing completely or were very desiccated. We did not remove any seedlings that emerged, nor did we clip back any vegetation from mature plants that encroached over the gap, as our aim was to simulate what occurred naturally. The seedling identification key of Chancellor (1966) and a seedling herbarium of all species present at the site (made by the authors) were used as aids in seedling identification.

Statistical analysis

Seed bank and seedling recruitment data were analysed in glim using log-linear models of counts (NAG 1985). As we conducted a large number of statistical tests, we a priori chose α = 0.01 as our level of significance. For the most frequently occurring species in the seed bank, we analysed the effect of the different herbivores (insects, molluscs and rabbits) and soil depth (0–8 cm vs. 8–16 cm of soil profile) on the number of germinable seeds that emerged from each soil sample. For the seedling recruitment data, we analysed for the most frequently occurring species the effect of different herbivores and seed source (‘seed bank + seed rain’ vs. ‘seed rain’) on the total number of seedlings that emerged into the gaps over the whole census period, the number of seedlings surviving in February 1997, and the proportion of seedlings that survived from autumn 1995 to February 1997. In each case an appropriate error structure ensured that residuals were constant and normally distributed. Poisson errors were used for germinable seeds and seedling counts, and binomial errors for the proportion surviving (Crawley 1993). For both the seed bank and seedling recruitment data, an entirely separate analysis was carried out for each level of the split-plot design, using total counts of germinable seeds or seedlings determined at the appropriate plot size. An empirical scale parameter was specified where the data were overdispersed (i.e. residual deviance substantially greater than residual degrees of freedom; Crawley 1993).

The association between the presence of a species in the seed bank and in the established vegetation, and the effect of different herbivores on the association, was analysed as a 6-dimensional complex contingency table in glim using a log-linear model with Poisson errors [i.e. present/absent in seed bank; present/absent in vegetation; with/without insecticide; with/without slug pellets; grazed/rabbit fenced; block 1/block 2; = 26 (64-cell) contingency table]. The same analysis was used to test the association between the species composition of the seedlings emerging in gaps and the vegetation.

Flowerhead data were analysed in glim using log-linear models with Poisson errors. Plant cover data were analysed by straightforward analysis of variance (anova) with normal errors, following arcsine square root transformation of the proportion.

Results

Composition of the vegetation

The point quadrat estimates showed a marked impact of rabbit fencing on the cover of species (Table 1). of the 11 species that were sufficiently widespread and abundant to carry out significance tests, four species had higher cover on rabbit-fenced plots, while two species (Galium saxatile and Rumex acetosella) had higher cover on rabbit-grazed plots. When interpreting these point quadrat results it is important to consider the marked impact rabbits had on vegetation structure. In particular, on fenced plots, where vegetation was tall in summer (e.g. 1996: sward surface height > 45 cm, mean standing biomass in summer = 463 g dry matter m–2; M.J. Crawley, unpublished data), some small creeping and short herbs at the base of the canopy (e.g. Galium saxatile and Ranunculus repens) may have been undetected. Furthermore, as only the first contact per species at each pin location was recorded, differences in the vertical distribution of a species were not accounted for, possibly leading to an underestimation of the proportional contribution of individual species on fenced plots. The point quadrat method used would be expected to give a more reliable estimate of the proportional contribution of individual species on unfenced grassland where rabbits kept the vegetation short for most of the year (e.g. 1996: sward surface height < 10 cm, mean standing biomass in summer = 291 g dry matter m–2; M.J. Crawley, unpublished data).

Table 1.  The main effects of exclusion of insects, molluscs and rabbits on plant cover (%) in early June 1995 and the number of germinable seeds that emerged (per m2) in 0–16-cm deep soil cores taken in late May 1995. Most (1552 out of 1706) of the seedlings that emerged were found in the top 8 cm of the soil profile. Arithmetic means without standard errors are presented as all tests of significance were done on transformed scales (arcsine for cover values and logs for seedling densities). The significance of the difference between the means is shown in the column to the right of each pair of means. (*P<0.01; NS, not significant. Other species did not occur frequently enough to be analysed)
 Vegetation (% cover)Seed bank (germinable seeds m−2)
 InsecticideMolluscicideRabbit fencingInsecticideMolluscicideRabbit fencing
SpeciesNoneSprayed NonePellets GrazedFenced NoneSprayed NonePellets GrazedFenced 
Only present in the vegetation
Achillea millefolium1.41.8 1.31.9 1.22.1 00 00 00 
Arrhenatherum elatius04.0 2.02.0 0.53.5 00 00 00 
Crepis capillaris02.4 02.4 0.61.8 00 00 00 
Festuca rubra30.538.1NS38.430.2NS28.839.4NS00 00 00 
Hieracium pilosella1.80 1.10.7 01.8 00 00 00 
Only present in seed bank
Hypochaeris radicata0.50.5 0.20.8 1.12.0 00 00 00 
Chenopodium album00 00 00 20.120.1 12.128.2 24.116.1 
Cytisus scoparius00 00 00 129.1112.9NS122.1119.8NS141.2100.8NS
Trifolium dubium00 00 00 28.04.0 28.04.0 24.28.1 
Ornithopus perpusillus00 00 00 60.540.3NS56.444.3NS56.444.4NS
Prunella vulgaris00 00 00 36.212.1 28.220.4 36.112.1 
Present in vegetation and seed bank
Agrostis capillaris14.98.2NS12.111.1NS6.217.1 * 1443.51241.9NS1318.51366.9NS887.11798.4 *
Anthoxanthum odoratum11.014.2NS13.112.1NS5.319.9 * 483.8584.6NS576.6491.9NS370.9697.6 *
Galium saxatile5.57.7NS5.67.6NS12.21.1 * 76.664.5NS60.580.6NS57.484.7NS
Holcus lanatus11.512.5NS12.710.9NS7.316.3 * 322.5370.9NS387.3306.5NS266.1427.4 *
Holcus mollis3.94.5NS3.84.4NS4.93.2NS8.124.2 16.116.1 15.915.9 
Leontodon autumnalis3.12.5NS3.02.6NS2.82.8NS8.116.1 8.116.1 4.120.1 
Lotus corniculatus1.93.5 3.42.0 0.94.5 60.572.5NS68.564.5NS52.479.6NS
Luzula campestris5.55.7NS4.76.4NS5.95.3NS20.224.2 20.224.2 16.128.2 
Plantago lanceolata4.25.7NS4.85.1NS2.07.8 * 64.5104.8NS76.692.7NS12.1157.3 *
Ranunculus repens00.5 00.5 0.50 412.1 4.112.1 12.14.1 
Rumex acetosa1.917.1 * 3.55.5NS2.56.5NS84.6125.0NS100.8108.8NS72.5137.1 *
Rumex acetosella2.53.5NS2.53.5NS5.50.5 * 76.656.4NS80.652.4 60.472.5NS
Senecio jacobaea1.91.3 2.01.2 2.40.8 31.815.9 15.931.8 31.815.9 
Stellaria graminea2.53.5 3.02.9 2.13.8 169.4165.4NS149.1185.5NS80.6254.0 *
Trifolium repens0.60.2 0.40.4 0.60.2 137.1145.2NS120.9161.2NS80.6201.6 *
Veronica chamaedrys1.41.6 1.02.0 1.21.8 32.320.2 36.316.1 20.232.6 
Vicia sativa0.50.7 0.30.9 0.50.7 189.5161.2NS185.5165.3NS60.4290.3 *
Total number of seedlings         3482.93398.5NS3472.43409.4NS2377.74502.8 *
Mean number of species17.618.4NS18.417.6NS18.817.2NS6.57.0NS6.06.9NS6.56.9NS

In contrast to the marked impact of rabbits, the effects of insect and mollusc exclusion on species cover were much less pronounced (Table 1). Rumex acetosa was the only species that had higher cover where insects were excluded, and this only occurred on rabbit-fenced plots (grazed-no insecticide = 1.0%, grazed-plus insecticide = 1.5%, fenced-no insecticide = 2.8%, fenced-plus insecticide = 7.1%). Application of slug pellets had no significant effect on the cover of any species. Insect and mollusc exclusion had little impact on vegetation structure (e.g. 1996: mean standing biomass (g dry matter m–2) in summer, sprayed = 401, unsprayed = 353, pellets = 367, no pellets = 392; M.J. Crawley, unpublished data).

There were no significant effects of insect, mollusc or rabbit exclusion on the number of plant species detected in the point quadrat data (glim with Poisson errors, P > 0.1).

The effect of the fertilizer, lime and plant competition treatments on the abundance of plant species from 1991 to 1997, and how this was modified by different kinds and intensities of herbivory, will be described elsewhere (M.J. Crawley & G.R. Edwards, unpublished data).

Seed bank composition and size

A total of 1706 germinable seeds of 22 plant species were found in the soil cores taken from Nash’s Field (Table 1). Most plant species (17) were present in both the seed bank and vegetation, with five species present only in the seed bank and six species present only in the vegetation. However, analysis of the data on the presence or absence of species in the seed bank and vegetation in each plot showed there was no close correspondence between the presence of a species in the seed bank and in the established vegetation (contingency table analysis, P > 0.1). Furthermore, there was no evidence that any of the herbivore exclusion treatments affected the number of species shared between the vegetation and seed bank (contingency table analysis, P > 0.1). The frequency distribution of germinable seeds was highly skewed, with three species, Agrostis capillaris, Anthoxanthum odoratum and Holcus lanatus, making up 39%, 16% and 10% of the total, respectively. Although dicots were the most common group, with 17 species found in the seed bank, germinable seeds of these species were few in number, making up only 35% of the total. The seed bank consisted primarily of perennial species, with the only annuals being Chenopodium album and Ornithopus perpusillus, both of which were absent from the above-ground vegetation.

Seeds were much more common in the top than bottom 8 cm of the soil profile; of the 1706 germinable seeds that emerged, 1552 (91%) were found in the top 8 cm. Only two species had an even distribution of seeds throughout the soil profile (Cytisus scoparius: 133.1 vs. 108.8 germinable seeds m–2 for 0–8 and 8–16 cm, respectively; Trifolium repens: 153.0 and 129.1 germinable seeds m–2 for 0–8 and 8–16 cm, respectively).

Rabbit fencing had a striking impact on the number of seeds in the seed bank (Table 1). The total number of germinable seeds was higher by a factor of 2 on rabbit-fenced plots, and of the 13 species that were common enough for significant tests, eight had more germinable seeds on rabbit-fenced plots. In contrast, insect and mollusc exclusion had no significant effect on the number of germinable seeds of any species. There were no significant effects of insect, mollusc or rabbit exclusion on the number of species of germinable seeds (glim with Poisson errors; Table 1), and no species was restricted to any one of the herbivory treatments.

Flowerhead production

The impact of rabbit fencing on flowerhead production was marked (Table 2). of the nine species that were common enough for statistical tests, seven had significantly more flowerheads on rabbit-fenced plots. As we did not make detailed measurements of flowerhead initiation, emergence and survival, it was not possible to say whether this effect was due to rabbits removing flowerheads (either as they developed or once developed) or due to grazing lowering the resources available for allocation to reproduction. Only Galium saxatile had significantly fewer flowerheads on rabbit-fenced plots. In contrast to the marked impact of rabbits, there were no significant effects of insect or mollusc exclusion on flowerhead density (Table 2).

Seedling recruitment into gaps

A total of 1913 seedlings of 26 species emerged in the gaps over the 22-month census period (Table 3). Most of these (1858) emerged in autumn (August to November) 1995, soon after the creation of the gaps. Dicots were the most common species in the gaps (20 of 26 species), but grasses (six of 26 species) were numerically the most abundant, making up 54% of the total seedlings that emerged. The grasses Holcus lanatus and Festuca rubra were the dominant species, making up 27% and 17% of seedlings, respectively. There was evidence of a positive relationship between the species composition of seedlings in gaps and the species composition of the established vegetation (contingency table analysis, P < 0.01) but no evidence that any herbivore exclusion treatment influenced the number of species shared between the gaps and the vegetation.

Table 3.  The effects of rabbit fencing on the total number of seedlings (means per m2 back-transformed from logs) that emerged into gaps over the whole census period (May 1995–February 1997), the number of seedlings surviving (means per m2 back-transformed from logs) in the gaps in mid-winter (February 1997), and the proportion (means back-transformed from logits) of seedlings that emerged in autumn 1995 that survived for over a year until mid-winter 1997. Means are averaged across insecticide, molluscicide and seed source (seed bank vs. seed bank + seed rain) treatments. The significance of the difference between the means is shown in the column to the right of each pair of means. *P<0.01, NS=not significant. All tests of significance were done on the transformed scales
 Number of seedlings emergedNumber of seedlings survivingProportion of seedlings surviving
SpeciesGrazedFenced GrazedFenced GrazedFenced 
  • Species analysed had at least 40 seedlings in total.

  • Species analysed had at least 30 seedlings in total surviving in gaps. Other species that emerged in gaps were (number of seedlings in parentheses): Arrhenatherum elatius (1), Achillea millefolium (3),Crepis capillaris (30), Cerastium fontanum (14), Cytisus scoparius (5), Holcus mollis (12), Hieracium pilosella (2), Hypochaeris radicata (11), Lotus corniculatus (24), Luzula campestris (3), Prunella vulgaris (1), Ranunculus repens (12), Rumex acetosella (36), Senecio jacobaea (8), Trifolium repens (7), Veronica chamaedrys (8), Vicia sativa (13).

Agrostis capillaris18.138.2 * 9.719.4 * 0.540.49NS
Anthoxanthum odoratum26.430.6NS13.29.7NS0.390.38NS
Festuca rubra118.1247.2 * 32.627.8NS0.250.12 *
Galium saxatile40.323.6 * 21.58.3 * 0.520.38NS
Holcus lanatus76.3161.8 * 23.629.9NS0.320.18 *
Leontodon autumnalis25.763.2 * 11.116.7NS0.360.23 *
Plantago lanceolata22.234.7NS9.012.5NS0.410.25 *
Rumex acetosa62.585.4 * 19.416.6NS0.340.20 *
Stellaria graminea49.376.3 * 11.813.9NS0.250.16NS
Total number of seedlings488.9839.6 * 164.5175.7NS   
Mean number of species9.610.3NS6.87.1NS   

The impact of rabbit fencing on seedling emergence was marked (Table 3). The overall number of seedlings that emerged was increased by a factor of 1.7 by rabbit fencing, and of the nine species that were common enough for analysis, six species had significantly more seedlings on rabbit-fenced plots. Only Galium saxatile had significantly fewer seedlings on rabbit-fenced plots. The total number of seedlings was not affected by the exclusion of insects (seedlings m–2: sprayed = 729.2, unsprayed = 599.3) or molluscs (seedlings m–2: pellets = 727.1, no pellets = 601.4). The number of seedlings of Rumex acetosa was higher on plots where insects were excluded (seedlings m–2: sprayed = 131.9, unsprayed = 16.0) and also higher on plots where molluscs were excluded (seedlings m–2: pellets = 97.9, no pellets = 50.0). It is worth noting that germination of Rumex acetosa was stimulated by insecticide application in glasshouse trials. For Stellaria graminea, there was a significant interaction between the effects of mollusc exclusion and rabbit fencing; seedling emergence was higher where molluscs were excluded but only on rabbit-fenced plots (seedlings m–2: grazed-no pellets = 35.6, grazed-pellets 49.0, fenced-no pellets = 50.6, fenced-pellets = 116.2 seedlings m–2). Insect, mollusc and rabbit exclusion had no significant effect on the number of species of seedlings emerging in the gaps.

There was little difference in seedling emergence between the ‘seed rain + seed bank’ and ‘seed rain’ treatments. The total number of seedlings that emerged was 629.9 m–2 in the ‘seed rain + seed bank’ treatment and 698.6 m–2 in the ‘seed rain’ treatment. Of the five species that were present only in the seed bank (Table 1), we observed just five seedlings of Cytisus scoparius. Furthermore, of the species present in both the seed bank and the vegetation (Table 1), only Galium saxatile had more seedlings in the ‘seed rain + seed bank’ treatment (39.6 seedlings m–2) than the ‘seed rain’ treatment (24.3 seedlings m–2). There was no significant difference between the ‘seed rain + seed bank’ and ‘seed rain’ treatments in the number of species of seedlings that emerged in gaps.

Seedlings surviving in february 1997

Few of the statistically significant treatment effects on seedling emergence were still significant when the number of seedlings surviving in February 1997 was analysed. The number of plants of Agrostis capillaris in February 1997 was still greater on fenced plots (Table 3), but there was no significant effect of fencing remaining for the other five species which had shown greater emergence on fenced plots (Table 3). Galium saxatile, which showed greater emergence on grazed plots, still had more seedlings on grazed plots in February 1997 (Table 3). The number of seedlings of Rumex acetosa in February 1997 was higher on plots where insects were excluded (seedlings m–2: sprayed = 30.6, unsprayed = 6.3) and also higher on plots where molluscs were excluded (seedlings m–2: sprayed = 25.0, unsprayed = 11.8). The interaction between the effects of mollusc exclusion and rabbit fencing for Stellaria graminea present for emergence was no longer significant. Galium saxatile seedlings were still more abundant in the ‘seed bank + rain’ (19.4 seedlings m–2) than the ‘seed rain’ treatment (10.4 seedlings m–2).

There was a marked impact of rabbit fencing on the proportion of seedlings surviving from autumn 1995 to February 1997 (Table 3). Survival of five species was lower on rabbit-fenced plots (Table 3). Most seedling mortality occurred during the rapid growth phase in spring (March–May) 1996. Insect and mollusc exclusion had no significant effect on the proportion of seedlings surviving.

Discussion

Recruitment from the seed bank and seed rain

Many field studies have shown that grasslands have large persistent seed banks (references in Thompson & Grime 1979), often with a different species composition to the above-ground vegetation, but few studies have addressed the extent to which these seeds recruit into the growing population (Sarukhan 1974; Bullock et al. 1994). In our study, the ‘seed rain + seed bank’ and ‘seed rain’ treatments were designed to differentiate between recruitment from the seed rain and seed bank, and showed that despite a large seed bank, few seedlings emerged from the seed bank into experimentally created gaps. Of the five species present only in the seed bank we observed just five seedlings of Cytisus scoparius, and of those species present in both the seed bank and vegetation only Galium saxatile had greater seedling densities in the ‘seed bank + seed rain’ treatment.

There are at least three possible explanations for why recruitment from the seed bank into our experimentally created gaps was negligible despite many seeds being present in the soil. First, because germination from the seed bank is only partially observable through those seedlings that successfully establish, we may have failed to detect seedlings that emerged from the seed bank but died during the period between successive seedling measurements. Secondly, although seeds were present predominantly in the top 8 cm of the soil profile, movement of seeds to the soil surface may have been insufficient for germination to occur. For instance, we removed perennial vegetation but did not disturb the soil surface or mix the soil profile, as might occur with natural disturbances such as rabbit scrapes and molehills. Without soil disturbance the specific conditions required for seed dormancy to be broken may not have been met (e.g. we observed few seedlings of Trifolium repens and Cytisus scoparius in the gaps but have observed many on molehills; G.R. Edwards, unpublished data). Thirdly, the timing of disturbance relative to the availability of seeds capable of germination may have been important (Hobbs & Mooney 1985; Lavorel et al. 1994; Kotanen 1996). For example, even if dormancy was broken by gap opening, the timing (e.g. late spring) may have been inadequate for successful germination (e.g. of autumn germinators) and could subsequently have resulted in the death of seeds. Furthermore, the timing of disturbance was close to the summer flush of seed dispersal, and rapid germination of cohorts of seedlings arising from the seed rain (e.g. Festuca rubra) may have reduced the germination probability of seeds in the persistent seed bank (Inouye 1980; Graham & Hutchings 1988; Rees & Brown 1991).

The finding that the seed bank played only a minor role in recruitment into experimentally created gaps in this grassland compared with the recent seed rain is similar to that described by Bullock et al. (1994) in a study of a sheep-grazed, species-poor grassland. This result suggests that the potential for the vegetation to change through recruitment in gaps from a seed bank with a different composition to the vegetation is low. An alternative way seedling recruitment in gaps may be important in vegetation change is if it had a different composition to the existing vegetation. This may occur when seed production is not proportional to shoot biomass (e.g. because of selective herbivory, species preferred by rabbits like Festuca rubra may produce little or no seed, despite making up a substantial proportion of the biomass). However, we found no evidence of this as the species composition of seedlings emerging in gaps was similar to that of the vegetation. Emergence was dominated by the same perennial grasses that were dominant in the vegetation (e.g. Festuca rubra) and no species novel to Nash’s Field were found that might have dispersed to the grassland (e.g. by wind or in rabbit faeces).

The conclusion that recruitment from the seed bank and seed rain is not important in vegetation change in Nash’s Field is specific to the gap treatments that we imposed. Recruitment from the seed bank and seed rain may be more important in vegetation change when larger scale gaps arise (e.g. disturbance by cultivation; Crawley 1990a) or in different types of vegetation gaps (e.g. patches of vegetation that differ in biomass or structure). In the latter case, the seed bank or seed rain need not differ in composition from the established vegetation for vegetation change to occur; studies have shown that turf and canopy gaps can provide particular environmental conditions that meet species-specific requirements for germination, so that certain species recruit within gaps regardless of the species composition of the seed bank or the seed rain (Rusch 1992; Van Tooren 1988; Rusch & Fernandez-Palacios 1995). It is also important to note that our results are based on an experiment conducted over the relatively short period of two field seasons (both of which had dry spring–summer periods compared with the long-term Silwood Park average; Edwards & Crawley 1999) and are based on defoliation treatments (i.e. rabbit grazing or annual hay crop) that prevented succession. In the longer term, where there is no rabbit grazing or mowing, recruitment from the seed bank and seed rain are clearly important in vegetation change as there is succession to Quercus robur–Betula pendula woodland (via the seed rain) or Cytisus scoparius thickets (via the seed bank) (M.J. Crawley, unpublished data).

It has been argued previously that species with large seed banks might be less likely to be seed limited [i.e. where an increase in seed production (seed rain) leads to an increase in seedling recruitment; Crawley 1990b] because seeds from the seed bank fill all the establishment microsites that come available (Crawley 1983). However, our data suggest that this need not always be the case as the seed bank may be closed to recruitment, and increases in seed rain (e.g. following herbivore exclusion) may well lead to increased seedling recruitment. Evidence to support this comes from a separate study in Nash’s Field, where we have found that four species that did form persistent seed banks (Plantago lanceolata, Rumex acetosa, Trifolium repens and Lotus corniculatus) were seed-limited; when seed of these four species was sown into the grassland, there was greater seedling recruitment than that occurring naturally from seeds arising from the seed bank and seed rain (G.R. Edwards, M.S. Heard & M.J. Crawley, unpublished data).

Effects of herbivore exclusion on the seed bank

This study is unusual in allowing us to assess the relative importance of different herbivores on the size and composition of the seed bank. Rabbit exclusion had a striking impact on the size of the seed bank, with eight species having greater seed banks on rabbit-fenced plots, whereas insect and mollusc exclusion had little impact. The finding of a marked impact of rabbit grazing is similar to that observed in other grasslands where reduced vertebrate grazing has resulted in an increase in the size of the seed bank (O’Connor & Pickett 1992; Bertiller 1996; see also Williams 1984), and the contrasting effects of rabbits compared to insects and molluscs probably reflects differences in their respective impacts on seed production. Although we did not measure seed rain directly (e.g. by seed traps), we found that rabbit grazing caused a massive reduction in flowerhead production, while invertebrates had negligible impact (Table 2; see also Plate 7 Crawley 1997).

The effects of insect, mollusc and rabbit herbivory on the number of plant species in the seed bank were negligible. There was no suggestion that insect, mollusc or rabbit herbivory directly excluded any plant species from the seed bank of this grassland, at least in the short term, and we did not find any species in the seed bank that were restricted exclusively to plots from which any particular herbivore was excluded. The closest thing to an excluded species was Plantago lanceolata (germinable seeds were found on only one rabbit-grazed plot but all rabbit-fenced plots). Otherwise, all species, including those that increased or decreased the most in the seed bank, were present both where herbivores were and were not excluded. Some species may be persisting in the grazed plots by virtue of long-lived seeds (e.g. Cytisus scoparius and Trifolium repens;Thompson & Grime 1979). Other species may be persisting because the seed bank is being replenished, albeit at a lower rate, by seeds dispersed to the plots from within (Table 2 shows that some flowerheads of each species were still recorded on grazed plots) or outside (i.e. by wind or in rabbit faeces) the experimental area. Moreover, the loss of species from the seed bank beneath this permanent grassland may be low because soil disturbance, and hence the opportunity to germinate and be grazed, is very low; we have found that a maximum of only 2–3% of the soil surface in this grassland is disturbed each year by rabbits and moles (G.R. Edwards & M.J. Crawley, unpublished data). The loss of species may take much longer and require higher rates of disturbance than that occurring at present in this grassland. Experiments are currently underway in disturbed (i.e. subject to annual cultivation) and undisturbed grasslands at Silwood Park to examine the long-term effects of rabbit grazing (> 10 years) on the loss of species from the seed bank, particularly those ones palatable to herbivores.

Effects of different herbivores on seedling emergence

Rabbit exclusion had a marked impact on seedling emergence, with six species having greater seedling densities on rabbit-fenced plots. Given our data it is difficult to say whether this result was due to greater seed rain (as predicted by flowerhead counts; Table 2), reduced seedling herbivory or a combination of these two processes. However, correlation analysis based on data from rabbit-fenced plots, only showed that for all six species there was a positive relationship between flowerhead production and seedling emergence (Spearman rank correlation, P < 0.05; Agrostis capillaris, Rs = 0.80; Festuca rubru, Rs = 0.92; Holcus lanatus, Rs = 0.81; Leontodon autumnalis, Rs = 0.74; Rumex acetosa, Rs = 0.73; Stellaria graminea, Rs = 0.86; flowerhead production of many of the species was so low on rabbit-grazed plots as to preclude correlation analysis). This result, in combination with the finding that far more flowerheads of these species were observed on rabbit-fenced plots (Table 2), suggests that the increased seed rain associated with rabbit exclusion was at least partly responsible for the higher emergence on fenced plots. Only Galium saxatile, which increased in cover under rabbit grazing, had enhanced seedling emergence on rabbit-grazed plots. Again, this may reflect the enhanced flowering of this species on grazed plots (Table 2), but it may also be due to reduced competition from the more vigorous grasses that were suppressed by rabbit grazing. To tease apart the mechanisms by which rabbits alter emergence in this grassland (e.g. seed rain, seedling herbivory and competitive release) requires seed-sowing experiments where seed rain is controlled, and these are currently underway in Silwood Park grasslands.

In contrast to the impact of rabbit exclusion, the effects of insect and mollusc exclusion on seedling emergence were negligible. Mollusc exclusion increased seedling densities of two species (Rumex acetosa and Stellaria graminea), and insect exclusion one species (Rumex acetosa). The reduction in seedling emergence of Rumex acetosa and Stellaria graminea by molluscs is similar to that described by Hanley et al. (1996) for seedlings derived from the seed bank. The reduction possibly reflects the medium to high palatability of these species to molluscs, coupled with their erect growth habit which makes them prone to grazing (Grime et al. 1968; Edwards & Gillman 1987). Consistent with Hanley et al. (1996), mollusc exclusion had no impact on seedling emergence in gaps of any grass species. Previous trials have shown that grasses such as Holcus lanatus and Agrostis capillaris may be unpalatable to molluscs (Grime et al. 1968). Grasses may also be grazing tolerant due to the presence of silica within leaf tissue (Cottam 1985) conferring protection to the seedling from mollusc herbivory.

To our knowledge the reduction in emergence of Rumex acetosa has not been published previously. As there is evidence of stimulation of seedling emergence by insecticide application for Rumex acetosa, the result may be an artefact of the pesticide application rather than a consequence of reduced insect herbivory. However, the difference in seedling densities in sprayed and unsprayed plots in the field was much larger than the germination stimulation we found in the glasshouse (field × 8.2; glasshouse × 1.5), and so it seems likely that exclusion of insects is important. This result possibly reflects pre- and post-dispersal seed predators that caused reductions in seed rain not evident from the flowerhead counts; we detected no difference in seedling survival between plots where insects were and were not excluded; and Hulme (1994a) found in the same grassland that insect herbivores of seedlings, such as lepidopteran larvae and tephritid flies, damaged seedlings but did not cause mortality.

Seedling survival

Whether differences in densities at the seedling stage ultimately affect plant populations depends on the extent to which there is compensatory density-dependent seedling survival. In our study there were some clear demonstrations that early differences did not persist, particularly for the effect of rabbit fencing. Emergence of six species was increased by rabbit exclusion, but when survival was considered this effect only remained for Agrostis capillaris (Table 3). The other species all showed a consistent pattern; rapid emergence in autumn 1995 following the hay cut, but considerable mortality, particularly on the fenced plots, in spring 1996. The higher mortality on the fenced plots was probably due to the rapid in-growth of competing perennial vegetation over the gaps in spring; survival may have been higher on the grazed plots because rabbits prevented biomass from accumulating. However, we cannot rule out that the response to fencing might have reflected indirect effects of fencing on other seedling mortality factors. For instance, Rice (1987) has demonstrated that small mammal abundance may in fact be higher within fenced exclosures, and their grazing effects may be confused with plant–plant competitive interactions. These findings confirm the views of Harper (1977) and Crawley (1997) that apparently important mortality factors that act at the seedling emergence stage can have negligible impact on plant recruitment and botanical composition because of density-dependent seedling survival at subsequent stages. This is particularly relevant given that many studies base conclusions about the importance of herbivores to plant recruitment on seedling emergence data alone (Edwards & Crawley 1999).

From this study it appears that the impact of exclusion of rabbits, insects and molluscs, individually or in combination, on seedling regeneration in gaps is small. As we did not include a treatment that excluded all of the herbivores present in the grassland, we cannot say whether other herbivores would have had any impact. For example, studies in Silwood Park grasslands have shown that rodents may have a considerable impact on seed populations (Hulme 1994b) and seedling survivorship (Hulme 1994a; Edwards & Crawley 1999), in the latter respect a more marked effect than insects or molluscs.

We conclude that under the disturbance and defoliation conditions of this experiment seedling recruitment from the seed bank was so low as to be almost undetectable. The species composition of the seedling populations in gaps was remarkably similar to that of the established vegetation and was dominated by the most abundant perennial grasses. This result suggests that the successful restoration of species richness in this grassland may require larger scale disturbances (e.g. rotavation or harrowing) in combination with seed sowing of species novel to the vegetation, followed by defoliation managements (e.g. cutting at different times) that prevent seed input from the dominant perennial grasses (Wells et al. 1989; Bullock et al. 1994).

Acknowledgements

The research is supported by a research grant from the Natural Environment Research Council. We thank Matt Heard, Phil Green and Gabre Asefa for help with collection of the data. We thank Mike Hay, Tony Parsons and Joanna Dixon for comments on earlier drafts. The manuscript benefited greatly from comments from Graciela Rusch and an anonymous referee.

Received 6 April 1998; revision accepted 2 November 1998

Ancillary