Influence of herbivory, competition and soil fertility on the abundance of Cirsium arvense in acid grassland

Authors


* Correspondence: Dr Grant Edwards, Department of Agriculture and Horticulture, TH Huxley School, Wye College, Wye, Ashford, Kent, TN25 5AH, UK.

Summary

1.  The extent to which the weed Cirsium arvense (creeping thistle) may be controlled by manipulating interspecific competition and herbivory was examined in two factorial experiments in order to identify non-chemical herbicide-based control methods for the weed.

2.  In the first experiment, a single spring cultivation of grassland intensively grazed by rabbits led to a 25-fold increase in C. arvense cover within 3 months, the effects of which were still present the following summer. As well as destroying the competing perennial vegetation, cultivation created and dispersed small root fragments (3–5 cm in length) from which almost all shoot recruitment occurred.

3.  Fencing the cultivated plots against rabbits decreased the cover of C. arvense because ungrazed regrowth from palatable/grazing intolerant species reduced recruitment of C. arvense seedlings and shoots. Seedling competition, in the form of a wildflower seed mix sown soon after cultivation, reduced C. arvense cover on fenced plots to pre-cultivation levels.

4.  In the second experiment, conducted in a permanent grassland, C. arvense shoot densities on plots fenced against rabbits and treated as a hay meadow were about one-eighth of those found on rabbit-grazed plots where competing vegetation was kept short. Adventitious shoot recruitment was greater on soil disturbances such as molehills and rabbit scrapes than in intact vegetation. Seedling recruitment occurred only on soil disturbances such as molehills.

5.  Lime and nitrogen fertilizer application to the fenced grassland increased the standing biomass of competing species, which reduced C. arvense shoot density. Outside the fences, rabbit grazing was so concentrated on the competing species of the nitrogen-fertilized and limed areas that C. arvense benefited from competitive release, exhibiting increased shoot density. Cirsium arvense showed pronounced competitive release from grasses, with greater shoot densities where grasses were removed with selective herbicides than where no plant species were removed.

6.  Exclusion of insects and molluscs with chemical pesticides had no effect on shoot or seedling recruitment or overall shoot density on cultivated soil or in permanent grassland.

7.  It is concluded that combinations of management procedures that encourage interspecific competition, such as sowing crops soon after cultivation and delaying grazing of them, and nitrogen fertilizer application and non- or reduced grazing of intact grasslands, will help reduce C. arvense abundance.

Introduction

Cirsium arvense (creeping thistle) (L.) Scop. is an important perennial weed of pasture, arable and conservation areas in the temperate regions of both hemispheres (Holm et al. 1977; Peel & Hopkins 1980; Donald 1990). It is a major problem as it restricts the area available for livestock grazing and reduces crop yields (Hartley & James 1979; O’Sullivan et al. 1982; references in Donald 1990). It is a particularly troublesome weed as new plants can recruit from both seed and small root fragments, and because one established plant can infest large areas by recruitment of shoots from adventitious buds on a creeping root system (Moore 1975; Grime, Hodgson & Hunt 1988).

In the management of C. arvense populations, chemical herbicide-based control measures are becoming increasingly unpopular, either because they are ineffective or uneconomic, or due to environmental concerns (e.g. effect on pasture legumes; Hartley & Thomson 1981), and non-herbicidal control methods are sought (Bourdôt et al. 1995). One approach that has been advocated is to enhance interspecific plant competition, so as to prevent the creation of conditions conducive to the recruitment of C. arvense shoots (Bourdôt 1996). The sowing of competitive smother crops is a method that utilizes this approach, and this has been shown to be effective in controlling C. arvense in some cases (Donald 1990). A further method is to manage interspecific competition through changes in the timing and intensity of grazing. The benefits of altering grazing management (e.g. lax vs. heavy grazing) will depend on the degree to which recruitment, growth and survival of C. arvense seedlings and shoots are inhibited by competing plants and how C. arvense is directly affected by grazing (e.g. seedling or shoot herbivory, removal of flowers and seeds; Amor & Harris 1975; Mitchell & Abernethy 1993). These demographic processes in C. arvense should, in principle, depend on the species and life-history stage (e.g. seedling vs. adult) of competing plants, as these will differ in their competitive ability (Thrasher, Cooper & Hodgson 1963; Hallgren 1976) and tolerance or palatability to herbivores relative to C. arvense (Crawley 1983).

A further approach advocated for the control of C. arvense is to enhance interspecific competition through fertilizer application (Donald 1990). Reductions in C. arvense abundance with fertilizer application might result from competing species being able to use and assimilate the added nutrients better, or from changes in the rate at which gaps appear and close in the vegetation (Bourdôt 1996). In studies to date, however, the response of C. arvense to fertilizer application has been inconsistent; for instance, nitrogen fertilizer has been shown to increase (Reece & Wilson 1983; Nadeau & Vanden Born 1990), decrease (Thrasher, Cooper & Hodgson 1963; Hume 1982) or have no effect (Hay & Ouellette 1959) on the abundance of C. arvense for reasons that are not clear (Donald 1990). Few studies have considered the effect of nutrients other than nitrogen (e.g. phosphorus and potassium; Donald 1990) or how fertilizer application might interact with grazing. There is increasing evidence from a wide range of plant communities that the grazing and diet selection behaviour of herbivores, and the impact that herbivores have on plant communities, alters as soil fertility changes along natural and artificial fertility gradients (Bazely 1990; Nams, Folkard & Smith 1996; John & Turkington 1997).

This paper reports on two factorial experiments conducted in rabbit-grazed acid grassland that examined the impact of herbivory and plant competition on shoot and seedling recruitment of C. arvense. Both experiments examined the effects of vertebrate and invertebrate herbivory on shoot and seedling recruitment by erecting rabbit fences and applying insecticides and molluscicides. The experiments differed, however, in how the effect of plant competition was examined. In the first experiment, the effect of competition from mature plants was examined using a cultivation treatment to remove the perennial cover, while the effect of competition from seedlings was examined by sowing a wildflower seed mix soon after cultivation. In the second experiment, the effect of different kinds and intensities of plant competition on C. arvense was examined by removing grass or herb species from the grassland with selective herbicides. Also in the second experiment, the effects of soil fertility and soil pH on C. arvense abundance were examined by applying nitrogen, phosphorus, potassium and magnesium fertilizers, and lime. Using these various treatments, we sought to test the idea that C. arvense abundance would be reduced in areas with increased growth or accumulation of competing vegetation (e.g. fenced vs. grazed, grassland vs. cultivated, sown vs. unsown, fertilized vs. non-fertilized) because of decreased recruitment of shoots and seedlings. We use our results to point out possible non-herbicidal control measures for C. arvense on recently cultivated soil and in intact grassland.

Materials and methods

Experimental sites

The study was carried out in two field experiments that were set up in species-poor mesic grassland on acid sandy soil at Silwood Park, Berkshire, UK (National Grid reference 41/944691; longitude 0°35′W, latitude 51°25′N) to examine the factors affecting the abundance of plant species (Edwards & Crawley 1999; Edwards, Crawley & Heard 1999). The experiments were conducted in two separate fields, known as Oak Mead (area 7 ha) and Nash's Field (area 6 ha), located 500 m apart but separated by an oak (Quercus robur L.) woodland. In both fields, the dominant grass species were Agrostis capillaris L. and Festuca rubra L. and the dominant herb species were Galium saxatile L. and Rumex acetosella L. (National Vegetation Classification: acidic variant of MG6; Rodwell 1992). Cirsium arvense was distributed throughout the grassland in both fields. It was generally present in low abundance (< 2% cover, < 5 shoots m−2) but some high-density patches existed (< 20% cover, > 50 shoots m−2). To our knowledge both fields had been under permanent grassland for at least the last 50 years. During this time, horses had periodically grazed the sites and hay crops had occasionally been taken. Both fields also had a long history of rabbit Oryctolagus cuniculus L. grazing; this kept standing biomass low for most of the year (< 150 g dry matter m−2) and precluded many species from flowering (Crawley 1990). Larger vertebrates observed grazing at the site were montjuc Muntiacus reevesi Oligby. and roe Capreolus capreolus L. deer. The vegetation cover was continuous except for small soil disturbances caused by the digging of rabbits and European moles Talpa europaea L.; < 3% of the ground area was disturbed each year (Edwards, Crawley & Heard 1999). Silwood Park experiences an average annual rainfall of 653 mm with little seasonal pattern. Additional information on the experimental sites can be found in Crawley (1990) and Edwards & Crawley (1999).

Oak mead experiment

Experimental design and treatments

The experiment was a five-factor factorial replicated in five 25 × 25-m blocks using a split-plot design. Half of each block was protected from vertebrate herbivores (mainly rabbits) and the other half left exposed. Each of these plots was split in three for interspecific plant competition treatments (cultivated, fumigated, intact grassland), and each of these, in turn, was split into four invertebrate herbivore exclusion treatments (with and without insecticides, and with and without molluscicides). Finally, in the centre of each invertebrate herbivory plot were two plots, which were assigned to a seedling competition treatment (wildflower seeds sown, no seeds sown). We used a large number of treatments in order to examine the relative and interactive effects on C. arvense of different kinds and intensities of competition and types of herbivores. Few previous experiments have been able to examine in a single experiment how different treatments might interact.

Vertebrate herbivory was manipulated by erecting wire mesh fences around one half (12·5 × 25 m) of each block in April 1996. These were 1-m high and constructed of 3-cm square wire mesh supported by posts every 4 m. The fences excluded rabbits (Crawley 1990) but not larger vertebrates like roe deer, which could easily jump the fences. Furthermore, the fences did not exclude moles, which tunnelled under the fences (Edwards, Crawley & Heard 1999), or small mammal seed predators (e.g. woodmouse Apodemus sylvaticus;Hulme 1994), which entered through the mesh. The fenced plots were left uncut in 1996 and 1997. In the remainder of the paper, plots grazed by rabbits will be termed ‘grazed’ and those fenced against rabbits termed ‘fenced’.

Three interspecific plant competition treatments, imposed by disturbing the intact grassland, were assigned to 8·3 × 12·5-m plots in each grazed and each fenced plot: (i)‘cultivated’, ploughed and rotavated to a depth of 25 cm in April 1996; (ii) ‘fumigated’, ploughed and rotovated in April 1996 followed by treatment with methyl bromide soil fumigant (gas applied beneath polythene sheeting for 7 days); (iii) ‘grassland’, intact undisturbed grassland.

The aim of these treatments was to create a range of levels of competition from the species naturally present in the grassland. The perennial cover on the intact grassland plots provided high competition; the regrowth from vegetative fragments and the seed bank on the cultivated plots provided medium competition; and the bare soil on fumigated plots provided low competition. Treatment of the soil with methyl bromide killed all of the vegetative fragments that survived the soil cultivation and all seeds in the seed bank except some hard-seeded legumes (e.g. Trifolium repens L. and Lotus corniculatus L.). Thus it provided close to a competition-free environment for germination of any seeds that dispersed to the plots. The fumigant also killed most, if not all, of the fungi and invertebrates in the soil (M.J. Crawley, unpublished data) and the confounding effects of this on the results must be borne in mind.

Invertebrate herbivory was manipulated by a factorial combination of chemical pesticides (with and without insecticide spray, and with and without molluscicide pellets) applied to four plots, each 6·25 × 4·1 m, in each interspecific competition plot. Insects were suppressed by a combination of knockdown (Ambush; Zeneca, Haslemere, Surrey, UK; cypermethrin synthetic pyrethoid at 150 g active ingredient ha−1) and systemic (Dimethoate-40; Atlas Interlades, Bradford, UK; dimethoate at 336 g active ingredient ha−1) insecticides. Molluscs were suppressed by applying pellets of metaldehyde (Mifaslug; Farmers Crop Chemicals Ltd, Inkberrow, Worcs, UK) at 960 g active ingredient ha−1. In 1996 and 1997, all pesticides were applied three times during the spring–summer period in April, May and June, and twice during autumn in September and October. Separate glasshouse trials gave no evidence that these pesticides, at the rates and frequencies used in the trial, had any effect on the germination and growth of C. arvense seedlings (G.R. Edwards, unpublished data).

Competition from seedlings was manipulated by sowing in April 1996 (after cultivation) 60 species of wildflowers into a 2 × 2-m plot in the centre of each invertebrate herbivore plot. Seeds were sown at a rate of 1000 seeds species−1 m−2. The species sown represented a range of herbs common to acid, calcareous and dry-sandy grasslands of the UK; no C. arvense seeds were sown. We considered the seed sowing to be analogous to crop addition after cultivation, or to over-sowing of grassland. Adjacent to each sown plot was a 2 × 2-m non-sown plot where recruitment of species from the seed bank and seed immigration was assessed.

This experimental design gave 240 plots in total; five rabbit fences, 30 interspecific plant competition plots and 120 invertebrate exclusion areas, each containing a sown and a non-sown 2 × 2-m plot.

Measurements

Cover. In August 1996 and 1997, visual estimates were made of the percentage ground cover of C. arvense over the entire 2 × 2-m area of each sown and non-sown plot. The estimates were made by two independent observers who agreed upon a single value for each plot. Values < 1% were scored as present; values > 1% and < 20% were scored to the nearest 1%; values > 20% were scored to the nearest 5%. On cultivated and fumigated plots, cover estimates were also made of all species other than C. arvense. As cover was estimated on a ground area basis it was not unusual on cultivated and fumigated plots for the sum of individual values for each species in any one 2 × 2-m plot to exceed 100%.

Shoot and seedling recruitment. In April 1996, immediately after cultivation, permanent 25 × 50-cm quadrats were established in the centre of each 2 × 2-m plot. The fate of C. arvense shoots and seedlings that emerged was followed by inserting toothpicks into the soil near each shoot or seedling and marking their location on a map of the quadrat. Seedlings were defined as having cotyledons present. At each census, seedlings and shoots were assigned as live or dead, and no effort was made to assign a cause of death. Censuses were done in April (mid-spring), July (mid-summer), October (mid-autumn) and January 1996 (mid-winter) and April and July 1997. From October 1996, shoots began to die through natural autumnal senescence, and by the end of December (mid-winter) there was no living tissue above ground. The extent to which above-ground shoots of C. arvense arose from root fragments compared with seeds was investigated by digging up 250 shoots on the cultivated plots in June 1996 and noting whether it was a seedling or a shoot from a root fragment.

Statistical analysis

All statistical analyses were performed using GLIM 3.77 (NAG 1985). The percentage cover of C. arvense in July 1996 and 1997 was analysed by split-plot analysis of variance (anova) following an arcsine transformation of the percentage. The number of shoots of C. arvense that recruited between censuses in the permanent quadrats was analysed by split-plot anova of √count + 1 transformed data. Four periods were analysed: April–July 1996 (late spring), July–October 1996 (summer), January–April 1997 (early spring) and April–July 1997 (late spring). In each case, the value analysed was the number of new shoots present at the end of each period. Shoot survival was analysed for two periods (July–October 1996 and April–July 1997) using log-linear models with binomial errors (Crawley 1993). Analysis was based on the number of shoots present at the beginning of the period that survived until the end of the period. Analysis of shoot recruitment and survival was only carried out for cultivated plots; insufficient shoots emerged on grassland and fumigated plots to make statistical analysis possible. The number of seedlings that recruited in two census periods (July–October 1996 and January–April 1997) was analysed by split-plot anova of √count + 1 transformed data. Analysis of seedling recruitment was only carried out for fumigated plots; insufficient seedlings emerged on the grassland and cultivated plots to make statistical analysis possible.

As fumigation also killed most, if not all, of the soil invertebrates (M.J. Crawley, unpublished data), insecticide and molluscicide treatments might be expected to have less (or no) impact on these plots, at least in the first year. Therefore, in the first year we restricted our analysis of the effect of insect and molluscs to grassland and cultivated (non-fumigated) plots only.

During fumigation, the plastic sheets used to retain the fumigant lifted before the completion of fumigation (due to strong wind) on parts of two blocks. The plots where this happened had the characteristics of the cultivated plots at the first census in July 1996; for example, there was evidence of recruitment of shoots from the seed bank and vegetative fragments that was absent on other fumigated plots. For the purposes of this study, we classified these plots as cultivated, and thus had unequal replication of cultivated and fumigated plots.

Nash's field experiment

Experimental design and treatments

This experiment, which began in 1991, was a six-factor factorial replicated in two blocks using a split-plot design (see Fig. 1 in Edwards, Crawley & Heard 1999). Invertebrate herbivore exclusion treatments (with and without insecticides, and with and without molluscicides) were assigned to four large plots in each block. Each of these large plots was split in half for exclusion of larger vertebrate herbivores (plus and minus rabbit fences), and each of these, in turn, was split into two soil pH treatments (limed and unlimed). Each of the liming treatments was split into three for the application of plant competition treatments (control, minus grass and minus herb species using selective herbicides). Finally, each of the plant competition plots was split into 12 plots for the application of mineral fertilizers (N, P, K and Mg). Like the Oak Mead experiment, we chose to use many treatments so that the relative and interactive effects on C. arvense of different kinds and intensities of competition, different types of herbivores and soil fertility could be examined within a single experiment. In addition, a large number of treatments (particularly nutrients) were used because of the desire to use the experimental resource to address a wide range of ecological questions (Crawley & Rees 1996; Edwards & Crawley 1999; Edwards, Crawley & Heard 1999).

Invertebrate herbivory was manipulated by a factorial combination of chemical pesticides (with and without insecticide sprays, and with and without molluscicide pellets) applied to the four large plots, each 22 × 44 m, in both blocks. Each plot was separated from others by at least 10 m. The chemicals used, and the rates and timing of application, were the same as those used in the Oak Mead experiment, except that pesticides were not applied in autumn. Rabbit grazing was manipulated by erecting fences around one half (plot size 22 × 22 m) of each invertebrate herbivory plot in June 1991. The fences had the same design as at Oak Mead. However, unlike the Oak Mead experiment the fenced plots were cut for hay with a hand-held sickle bar mower in late August each year. By this time C. arvense had finished flowering and had dispersed seed (M.S. Heard, unpublished data). All of the cut herbage was raked and removed from the plots to prevent the accumulation of dead organic matter.

Soil pH was manipulated by applying lime (CaO) at 20 tonnes ha−1 to an 8 × 18-m plot in each fenced and each grazed plot in autumn 1991 and again in autumn 1994. One further 8 × 18-m plot was left unlimed in each grazed and fenced plot. There was a 2-m guard strip around the outside of each plot and a 2-m gap between the limed and unlimed plots. Soil pH before the experiment started was 4·1. This is a low value for grassland and suggests increased growth of vegetation would occur following lime application. Soil pH on limed plots in summer 1997 was 7·0.

Three plant competition treatments were assigned to 6 × 8-m plots within each limed and non-limed plot: (i) ‘control’, intact grassland; (ii) ‘minus-herb herbicided’, herbs removed with selective herbicide (Pasturol; Farmers Crop Chemicals, Inkberrow, Worcs, UK; dicamba + MCPA + mecoprop at 2·7 kg active ingredient ha−1) applied in late April each year from 1992 to 1994 inclusive; (iii) ‘minus-grass herbicided’, grasses removed with selective herbicide (Clout; Hortichem, Ongar, Essex, UK; alloxydim-sodium at 1·0 kg active ingredient ha−1; or Checkmate; Rhone Poulenc Agriculture Ltd, Ongar, Essex, UK; sethoxydim at 870 g active ingredient ha−1) applied in late April each from 1992 to 1994 inclusive.

These treatments created marked differences in the proportion of grasses, herbs and C. arvense present in the vegetation. In 1994, a year after the herbicides were applied, the proportional contributions of grasses, herbs and C. arvense to the total shoot weight on control plots were 85·1, 13·6 and 1·3%, respectively (M.J. Crawley, unpublished data). Minus-herb plots were grass rich (98·4%), with little C. arvense (< 0·1%) or other herb species (1·5%). The minus-grass plots still contained significant quantities of grass (61·0%; F. rubra appeared to be resistant to the herbicides) but were herb rich (37·1% herbs and 1·9% C. arvense). Our rationale in using the minus-grass treatment was to assess the importance of interspecific competition from grasses in determining C. arvense abundance. If grass competition is important, C. arvense should increase compared with controls on the minus-grass herbicided plots. The minus-herb plot provided a comparison with chemical herbicide methods of control for C. arvense.

Each competition plot was divided into 12 plots each measuring 2 × 2 m, arranged in three columns of four, without guard strips. Four fertilizer nutrients [nitrogen (N) in the form of ammonium nitrate, potassium (K) as muriate of potash, phosphorus (P) in the form of triple superphosphate, and magnesium (Mg) as Epsom salts] were applied at the following rates: N at 150 kg ha−1, P at 35 kg ha−1, K at 225 kg ha−1 and Mg at 11 kg ha−1 during the first week in April each year from 1992 to 1997 inclusive. Fertilizer treatments were applied in pairs to adjacent plots, with each pair allocated at random, independently for each competition plot. The pairs of nutrient treatments were applied as follows: N or everything-but-N (i.e. P, K and Mg); K or everything-but-K (NPMg); P or everything-but-P (NKMg); Mg or everything-but-Mg (NPK); P and K together; N and Mg together; all nutrients (NPKMg); no nutrients. The rationale for applying the nutrients in pairs, rather than in a fully randomized design, was twofold: to minimize the risk of applying nutrients to the wrong plot, and to give an immediate visual impression of the role of each nutrient in adjacent plots. The soils of Nash's Field are characterized by low nitrogen and phosphorus status (M.J. Crawley, unpublished data). Measurements of P made in 1991 [5·6 ± 0·47 (1 SE, n = 8) mg kg−1 NaHCO3 soluble P] were well below the recommendation for growing grass and silage (> 20 mg P kg−1). Levels of exchangeable potassium [88·4 ± 7·0 (1 SE, n = 8) mg kg−1] and magnesium [37·5 ± 21 (1 SE, n = 8) mg kg−1) were also medium to low and may have been limiting growth (McLaren & Cameron 1990).

This experimental design gave 1152 plots in total; eight invertebrate herbivory plots, 16 rabbit grazing plots, 32 lime plots and 96 competition plots, each containing 12 2 × 2-m fertilizer plots.

Measurements

Shoot density. In July 1996 and 1997, all visible shoots of C. arvense were counted in each 2 × 2-m fertilizer plot.

Shoot and seedling recruitment. To examine whether small-scale soil disturbances caused by burrowing animals acted as locally important recruitment sites, we compared the recruitment and survival of seedlings and shoots of C. arvense on molehills and rabbit scrapes with areas of intact vegetation. In July 1995, permanent 50 × 50-cm quadrats were established on the disturbed soil areas of 30 molehills (15 in grazed plots and 15 in fenced plots) and 30 rabbit scrapes that had formed in the experiment over the early part of the summer. Scrapes were defined as areas of bare ground > 0·01 m2 where rabbit digging had exposed bare soil and where the soil had been excavated (average area of individual scrape = 0·04 m2). Molehills were defined as areas of loose soil > 0·1 m2 that had been pushed up on the soil surface and which covered the existing vegetation (average size of individual molehill = 0·14 m2). Permanent 50 × 50-cm quadrats were also established in areas of intact vegetation in the same fertilizer plot that the rabbit scrapes or molehills were in. To examine whether the timing of soil disturbance affected recruitment of C. arvense shoots, the procedure was repeated in April 1996 with a different set of 30 molehills (15 in fenced plots, 15 in grazed plots) and 15 rabbit scrapes that had formed during the winter and early spring period. Quadrats in intact vegetation in the same plot were again established. The fate of C. arvense seedlings and adventitious shoots that emerged was followed using the same procedure as in the Oak Mead experiment. Censuses were taken July and October 1995, January, April, July and October 1996, and January and April 1997.

Statistical analysis

The number of visible shoots in July 1996 and 1997 was analysed using log-linear models with Poisson errors (Crawley 1993). Analysis was based on shoot numbers summed at each plot size of the split-plot design. Where statistical models were over-dispersed (i.e. residual deviance greater than the residual degrees of freedom), a scale parameter was estimated empirically (Crawley 1993) and F-tests rather than χ2 tests were conducted. The correction for over-dispersion was done at all levels of the split-plot design except the level of nutrient treatments. The number of shoots and seedlings of C. arvense that recruited between censuses on the disturbed soil of molehills and rabbit scrapes, and in areas of intact vegetation, was analysed using log-linear models with Poisson errors. For summer disturbances, five periods were analysed: July–October 1995, January–April 1996, April–July 1996, July–October 1996 and January–April 1997. For winter disturbances, the last three time periods listed were analysed. In each case, the value analysed was the number of new shoots present at the end of each period.

Results

Oak mead experiment

Cover

The cover of C. arvense was dependent on the plant competition × rabbit grazing × seed sowing interaction in both 1996 (F2,99 = 6·48, P < 0·01) and 1997 (F2,99 = 4·57, P < 0·05) (Fig. 1). There were no significant main effects of the insecticide or molluscicide treatments in 1996 (insecticide: F1,48 = 0·006, P > 0·5; molluscicide: F1,48 = 0·001, P > 0·5) or 1997 (insecticide: F1,69 = 0·004, P > 0·5; molluscicide: F1,69 = 0·001, P > 0·5) and none of the interactions involving insecticide and molluscicide treatments was significant. On grassland plots in July 1996, C. arvense cover was low (< 2·0%) and there were no significant effects of rabbit grazing or seed sowing. In July 1997, C. arvense cover on grassland plots was reduced by fencing but there was no effect of seed sowing. All cultivated plots in July 1996, 3 months after ploughing and rotavation, had higher cover values than the grassland plots, except those where seeds were sown inside the fences (Fig. 1). Cover was highest on the grazed plots, and on these plots seed sowing had no effect. In contrast, on the fenced cultivated plots seed sowing reduced the cover of C. arvense. The effect of the rabbit grazing × seed sowing interaction was still significant on cultivated plots in July 1997 even though there was no repeat cultivation in April 1997. On fumigated plots in July 1996 there was little C. arvense: it was present on eight out of 73 plots but always in low abundance (< 1%; Fig. 1). By July 1997, C. arvense was present on 47 out of 73 plots and had a greater cover on the grazed plots. Seed sowing had no significant effect on the cover of C. arvense on fumigated plots in July 1997.

Figure 1.

The effect of rabbit grazing and the sowing of a wildflower seed mix on the mean percentage cover of Cirsium arvense in July 1996 and 1997 in three different interspecific competition treatments (grassland, cultivated and fumigated) in the Oak Mead experiment. White bars = not sown; black bars = sown. Values have been back-transformed following analysis of arcsine-transformed data and are averaged across the insecticide and molluscicide treatments. Bars with different letters are significantly different according to a LSD test (P = 0·05) on the transformed scale. Cultivation, fumigation and seed sowing were carried out in April 1996.

Shoot recruitment

Recruitment and survival of shoots was strongly affected by interspecific plant competition. On grassland plots, where there was a closed canopy, few adventitious shoots emerged in any of the census periods (36 shoots in total, maximum number of shoots recruited in any 3-month census period = 2·4 m−2), irrespective of seed sowing or grazing treatments. In contrast, many shoots (575 in total) emerged on cultivated plots, where the perennial cover had been destroyed (Table 1). Most of these shoots, at least in the first season after cultivation, arose from small root fragments. Of the 250 shoots dug up in spring 1996, 233 were attached to a root fragment 3–5 cm in length that had been chopped up by the rotavator; the remainder were seedlings. On cultivated plots, shoot recruitment and survival were greater on grazed plots, where rabbits reduced interspecific competition by eating regrowth of competing species arising from root fragments and seed (Tables 1 and 2). Shoot recruitment and survival were reduced by seed sowing on the cultivated fenced plots, where seed sowing led to an increase in the cover of competing species (Table 2). In contrast, on grazed plots seed sowing had no effect on the cover of competing species (Table 2), and there was no effect of seed sowing on shoot recruitment and survival (Table 1). There were no significant effects of the insecticide or molluscicide treatments on the number of shoots that recruited on cultivated plots. On fumigated plots, only four shoots were detected in spring 1996, thus demonstrating that fumigation effectively killed most fragments and seed. The few shoots that did emerge may have arisen from root fragments dispersed to the plots by water or animals, or from roots deep in the soil that were unaffected by fumigation (Donald 1990). Insufficient shoots emerged on fumigated plots throughout the study to make statistical analysis possible, although the observed pattern was increased recruitment from one census period to the next and greater recruitment on grazed plots. For example, in the last census period (April–July 1997) the mean number of shoots that emerged was 4·4 m−2 on grazed plots and 0·4 m−2 on fenced plots.

Table 1.  The effect of rabbit grazing and the sowing of a wildflower seed mix on (a) the mean number of Cirsium arvense shoots recruited m−2 and (b) the proportion of shoots surviving on the cultivated plots in the Oak Mead experiment. Recruitment values are the number of new shoots present at the end of the measurement period, and so represent the emergence of shoots in a period minus the death, before the end of the period, of shoots that emerged. Recruitment values are arithmetic means averaged across insecticide and molluscicide treatments. Survival values are the proportion of shoots surviving from the start to the end of each period. Survival values are means back-transformed following analysis on the logit scale and are averaged across insecticide and molluscicide treatments. The seed sowing × rabbit grazing interaction was significant in all cases. **P < 0·01, *P < 0·05. Values with different letters within each column are significantly different according to a LSD (P = 0·05) test on the transformed scales (√count + 1 for number of shoots recruited; logit for survival)
  Late spring 1996
(April–July)
Summer 1996
(July–October)
Early spring 1997
(January–April)
Late spring 1997
(April–July)
(a) Recruited shoots
GrazedNot sown56·6 a23·2 a36·3 a35·2 a
GrazedSown61·0 a26·1 a38·4 a37·8 a
FencedNot sown24·0 b9·4 b16·0 b15·2 b
FencedSown6·1 c4·5 c5·3 c7·5 c
Seed sowing × rabbit grazing interaction (F1,39)
(b) Survival
 7·9 **
4·2 *
4·6 *
5·2 *
 
GrazedNot sown 0·65 a 0·53 a
GrazedSown 0·63 a 0·58 a
FencedNot sown 0·45 b 0·36 b
FencedSown 0·25 c 0·27 c
Seed sowing × rabbit grazing interaction (χ2, 1 d.f.) 3·9 * 4·1 * 
Table 2.  The effect of rabbit grazing and seed sowing on the percentage ground cover of species excluding Cirsium arvense on cultivated and fumigated plots in the Oak Mead experiment. Values are averaged across insecticide and molluscicide treatments. There was a significant plant competition × rabbit grazing × seed sowing interaction in each year (split-plot anova 1996: F1,67 = 7·27; 1997: F1,67 = 6·5, P < 0·01). Values with different letters within each year are significantly different according to a LSD (P = 0·05) test. Total cover on grassland plots was not assessed; however, bare ground on these plots was always < 1%
Plant competitionRabbit grazingSeed sowingJuly 1996July 1997
  • *

    Cover higher than expected, particularly in the first year, because Lotus corniculatus and Trifolium repens recruited from seeds in the seed bank that were not killed by soil fumigation.

CultivatedGrazedNot sown54·5 c83·4 c
GrazedSown56·8 c96·1 c
FencedNot sown82·1 b116·1b
FencedSown98·3 a137·1a
FumigatedGrazedNot sown1·9 e65·2 d
GrazedSown12·9 d90·2 c
FencedNot sown*79·4 b111·4 b
 FencedSown103·4 a146·7 a

Seedling recruitment

No seedlings were found on grassland plots during the whole experiment. Similarly, few (23 in total) seedlings emerged on cultivated plots where there was competition from regrowth arising from fragments and the seed bank. The only significant seedling recruitment occurred on fumigated plots (74 seedlings in total), and here it was almost entirely restricted to grazed plots (71 seedlings) in two periods: summer 1996 (35 seedlings) and early spring 1997 (39 seedlings). Presumably these were seedlings that arose from seeds dispersed to the fumigated plots in summer 1996 from C. arvense plants that flowered on the surrounding grassland and cultivated plots. In these periods the soil had remained nearly bare due to rabbits removing the competing seedlings that emerged. On the grazed plots, there were no significant effects of seed sowing on seedling recruitment in summer 1996 (sown = 12·8 seedlings m−2, non-sown = 16·8 seedlings m−2, F1,14 = 2·0, P > 0·1) or in early spring 1997 (sown = 14·4 seedlings m−2, non-sown = 19·2 seedlings m−2, F1,14 = 1·7, P > 0·1). The failure of C. arvense seedlings to emerge on grazed fumigated plots after early spring 1997 might reflect the growth on these plots in spring 1997 of some non-sown unpalatable species (e.g. Senecio jacobaea L.) that arose from seeds dispersed to the plots from the surrounding grassland the previous summer (M.J. Crawley & G.R. Edwards, unpublished data). It is perhaps surprising that few seedlings emerged in autumn 1996 and spring 1997 on fumigated plots inside the fences, even under the benign experimental conditions of the non-sown plots. This probably reflects the high level of competition on these plots, which arose because Trifolium repens and Lotus corniculatus recruited from seeds in the seed bank that were not killed by the soil fumigation. Both species were highly palatable to rabbits, so by July 1996 the sown parts of the grazed plots were nearly completely bare soil, whereas the fenced plots were covered by an unbroken carpet of T. repens, with occasional individuals of L. corniculatus (Table 2). The insecticide and molluscicide treatments had no significant effect on the number of seedlings that emerged on fumigated plots in autumn 1996 (insecticide: F1,10 = 0·7, P > 0·1; molluscicide: F1,10 = 0·3, P > 0·1) or in early spring 1997 (insecticide: F1,10 = 0·001, P > 0·1; molluscicide: F1,10 = 0·001, P > 0·1).

Nash's field experiment

Shoot density

Shoot density was strongly affected by rabbit grazing in 1996 (F1,4 = 45·1, P < 0·01) and 1997 (F1,4 = 36·9,P < 0·01). Shoot density on rabbit fenced plots was about 12% that of rabbit-grazed plots in both years. There was a significant rabbit grazing × lime interaction in 1996 (F1,8 = 25·1, P < 0·01) and 1997 (F1,8 = 15·8, P < 0·01). On fenced plots where shoot density was low, lime application decreased shoot density to less than 0·01 shoot m−2. On grazed plots, lime application increased shoot density by a factor of four in 1996 and a factor of two in 1997. Shoot density was greater on the minus-grass herbicided plots than the controls plots on both grazed and fenced plots in 1996 (F2,32 = 9·4, P < 0·01) and 1997 (F2,32 = 8·6, P < 0·01). There were few shoots on the minus-herb herbicided plots 3 years after herbicide application (Fig. 2). There were no significant effects on shoot density in either year of the application of insecticide (1996: none = 1·21 shoots m−2, applied = 1·52 shoots m−2, F1,3 = 3·5, P > 0·05; 1997: none = 1·12 shoots m−2, applied = 0·84 shoots m−2, F1,3 = 3·8, P > 0·05) or molluscicide (1996: none = 1·35 shoots m−2, applied = 1·44 shoots m−2, F1,3 = 0·51, P > 0·1; 1997: none = 0·98 shoots m−2, applied = 0·95 shoots m−2, F1,3 = 0·87, P > 0·1).

Figure 2.

The effect of (a) lime application and (b) selective plant species removal on the mean number of Cirsium arvense shoots m−2 in July 1996 and 1997 on rabbit-grazed and rabbit-fenced plots in the Nash's Field experiment. Values have been back-transformed from logs following analysis at each plot size of the split-plot design using GLIM with Poisson errors. Shoot density was affected by the lime × rabbit grazing interaction in both years. Shoot density increased with grass removal on both grazed and fenced plots. Bars with different letters within each year are significantly different according to a LSD (P = 0·05) test on the log scale.

Nitrogen application increased shoot density on grazed plots but decreased shoot density on fenced plots (rabbit grazing × nitrogen interaction; 1996: χ2 = 3·9; 1997: χ2 = 4·1; both d.f. = 1 and P < 0·05) (Fig. 3). Phosphorus application increased shoot density on both grazed and fenced plots (1996: χ2 = 8·6; 1997: χ2 = 4·3; both d.f. = 1 and P < 0·01) (Fig. 3). There were no significant effects of potassium (1996: χ2 = 0·8; 1997: χ2 = 3·2; both d.f. = 1 and P > 0·50) or magnesium (1996: χ2 = 1·1; 1997: χ2 = 2·8; both d.f. = 1, P > 0·5) application on shoot density (Fig. 3). None of the interactions between nutrient treatments were significant.

Figure 3.

The effect of (a) nitrogen, (b) phosphorus, (c) potassium and (d) magnesium fertilizer application on the mean number of Cirsium arvense shoots m−2 in July 1996 and 1997 on rabbit-grazed and rabbit-fenced plots in the Nash's Field experiment. White bars = no fertilizer; black bars = fertilizer applied. Values have been back-transformed from logs following analysis using GLIM with Poisson errors. Shoot density was affected by the nitrogen × rabbit grazing interaction in both years. Shoot density increased with phosphorus application on both grazed and fenced plots in each year. Bars with different letters within each year are significantly different according to a LSD (P = 0·05) test on the log scale.

Shoot and seedling recruitment

Recruitment of seedlings and shoots was far greater on grazed than fenced plots (total seedlings on molehills: grazed = 78, fenced = 3; total shoots on molehills: grazed = 266, fenced = 8). On grazed plots, seedlings were only found on disturbed sites and occurred in greatest numbers in the periods immediately after the disturbance was formed; no seedlings were found once competing vegetation had encroached over the soil (Table 3). Shoot recruitment was greater on disturbed sites than in intact vegetation at the censuses soon after soil disturbance (Table 3). At later censuses, when competing vegetation had encroached over the gap, there was no difference in shoot recruitment between disturbed sites and intact vegetation. There was little difference in seedling and shoot densities between early summer and winter/spring disturbances, or between rabbit scrapes and molehills (Table 3).

Table 3.   The number of seedlings and shoots (per m2) that recruited on the disturbed soil surface of molehills and rabbit scrapes (D) and in intact vegetation (V) on rabbit-grazed plots in the Nash's Field experiment. Values are the number of new shoots present at the end of census period, and so represent the emergence of shoots in a period minus the death, before the end of the period, of shoots that emerged. The arithmetic mean for each treatment is shown and the significance of the difference in shoot densities between disturbed sites and intact vegetation calculated by GLIM analysis with Poisson errors is shown in the right hand column for each comparison. **P < 0·01, *P < 0·05, NS, non-significant. There were too few seedlings for statistical analysis
Summer 1995 (July–October)Early spring 1996 (January–April )Late spring 1996 (April–July)Summer 1996 (July–October)Early spring 1997 (January–April)
Disturbance
formed
Disturbance
type
DVDVDVDVDV
Early summer 1995Molehills          
Seedlings8·603·60000000
Shoots16·04·3 **10·15·3 *7·24·8 NS9·25·8 NS5·17·2 NS
Rabbit scrapes          
Seedlings7·801·90000000
Shoots17·34·0 **9·45·1 *5·64·8 NS6·15·8 NS5·15·3 NS
Winter/spring 1996Molehill          
Seedlings    6·901·6000
Shoots    14·13·6 **5·64·8 NS4·02·7 NS
Rabbit scrapes          
Seedlings    7·501·903·60
Shoots    14·45·1 *6·23·6 NS7·25·6 NS

Discussion

One approach that has frequently been advocated for weed management is to reduce the recruitment and growth of weed species by enhancing interspecific plant competition (Altieri et al. 1987; Berkowitz 1988; Donald 1990). Our study has shown that grazing, soil disturbance, soil fertility and seed sowing all affected the level of interspecific competition experienced by C. arvense which, in turn, influenced C. arvense abundance by determining the extent of recruitment of C. arvense seedlings and adventitious shoots. It was possible to identify combinations of these factors that led to reduced abundance of C. arvense in perennial grassland and on recently cultivated soil.

Effect of cultivation on shoot recruitment

There was a striking increase in the cover of C. arvense on rabbit-grazed plots following a single soil cultivation. Cover increased from less than 3% to greater than 20% within 3 months. This result is consistent with early reports (see References in Donald 1990) that suggest cultivation can spread and increase the abundance of C. arvense. Cultivation not only destroyed the competing vegetation, creating competition-free sites for seedling and shoot recruitment, but also created and dispersed small root fragments (3–5 cm in length) from which adventitious shoots recruited (Sagar & Rawson 1964). Two of our results, however, indicate that interspecific competition could prevent this marked increase in abundance following cultivation.

First, the abundance of C. arvense was lower on fenced cultivated plots, where competing ungrazed vegetation (arising from root fragments, seed bank and sown seeds) accumulated, than on grazed plots, where rabbits kept the competing vegetation short. On grazed plots, C. arvense appeared to benefit from release from competitors as a result of the associated vegetation being more preferred by rabbits or less tolerant to grazing. We know of no other data on the effect of grazing on C. arvense shoot recruitment on recently cultivated soil. However, the result is consistent with the observation that newly established forages on cultivated soil suppressed C. arvense better when initial mowing was delayed until the second year after cultivation (Derscheid, Nash & Wicks 1961; Donald 1990).

Secondly, on fenced cultivated plots the abundance of C. arvense was reduced to grassland levels where wildflower species recruited from sown seeds. This result is consistent with other studies that have noted that recently sown crops can reduce C. arvense abundance (Hallgren 1976; references in Donald 1990). The important point is that seedlings of the sown wildflowers (e.g. Leucantheum vulgare Lam., Achillea millefolium L. and Reseda lutea L.) not only prevented almost all C. arvense seedling recruitment but also reduced recruitment of adventitious shoots from root fragments. This contrasts with the general view that shoots recruiting from storage organs, such as the roots of C. arvense, are more competitive than seedlings (Howe & Snaydon 1986; Crawley 1990). This result highlights that early canopy closure, even of seedlings of species considered inferior competitors (i.e. a species that would lose in adult to adult competition or seedling to seedling competition), is an important factor that can inhibit C. arvense shoot recruitment.

Effect of local disturbances on seedling and shoot recruitment in otherwise intact grassland

Heimann & Cussans (1996) concluded that seedling recruitment of C. arvense was rare, being restricted to disturbed habitats such as arable cropping land. Our results are consistent with this. In Nash's Field seedling recruitment was restricted to molehills and rabbit scrapes, and in Oak Mead there was considerable seedling recruitment on the competition-free seed bed of fumigated grazed plots in autumn 1997 but little in spring 1997 once the area had been colonized by unpalatable species (e.g. Senecio jacobaea). We also found in intact grassland that shoot recruitment was much greater in disturbed sites. Bakker (1960) attributed low seedling recruitment of C. arvense in intact grassland to the light requirements for germination not being fulfilled. Light probably plays no role, however, in recruitment of adventitious shoots on disturbances, as buds are able to grow to the soil surface from depths at which light does not penetrate. These shoots emerge to become new recruits once winter frosts have finished (G.W. Bourdôt, unpublished data). It is more probable that the adventitious shoots on molehills and rabbit scrapes arose either from root fragments caused by the animals’ digging or from the axillary buds on the below-ground part of shoots with damaged growing points. Some small root fragments were observed on rabbit scrapes in dense stands of C. arvense, indicating that natural disturbances could potentially act as direct analogues to soil cultivation.

Effect of interspecific competition in grassland

Thrasher, Cooper & Hodgson (1963) and Ang et al. (1994), among others, have hypothesized that the abundance of C. arvense in grassland is strongly affected by the level of interspecific competition from the established grassland sward, particularly perennial grasses. However, existing experimental evidence for this effect is poor (reviewed by Bourdôt 1996). Three of our results, however, provide some support for the hypothesis.

First, on fenced plots in the Nash's Field experiment, where ungrazed biomass accumulated (Crawley 1990; Crawley & Rees 1996), there was a marked reduction in the overall density of C. arvense shoots and the recruitment of new shoots and seedlings. It is worth noting, however, that the response to fencing may be a slow process. For instance, in the Oak Mead experiment, fencing decreased the cover of C. arvense only in the second year. A further factor, in addition to reduced competition, that may have contributed to the increase in C. arvense shoot density on grazed plots is the disturbance regime. Not only were disturbed sites on grazed plots better sites for shoot and seedling recruitment than those on fenced plots, but the area disturbed on grazed plots was also greater [total area disturbed (m2) by molehills and rabbit scrapes per m2 of ground area during 1995 and 1996: grazed = 0·051, fenced 0·038]. When considering the role of vertebrate grazers, it is important to note that our study used rabbits as the vertebrate grazers, whereas the grazers on farms and nature reserves are generally livestock. As livestock differ from rabbits in their size, the level of physical disturbance caused and grazing preference, their impacts may be different. Indeed, several studies have shown hard grazing by sheep and cattle, either intermittent or continuous, reduces patch expansion and shoot density in C. arvense compared with lax grazing (Amor & Harris 1975; Hartley & Thomson 1981; Hartley, Lyttle & Popay 1984; Mitchell & Abernethy 1993). It seems that in these studies, the reduced interspecific competition usually associated with increased grazing is, under very heavy grazing pressure, overshadowed by an opposing force, that of being eaten or trampled (Mitchell & Abernethy 1993). Thus, using ‘hard’ grazing by livestock may be an equally effective way of controlling the thistle as no grazing. This is supported by previous studies showing that multiple defoliations within a growing season (by mowing) reduce C. arvense abundance by reducing the leaf area duration and root reserves (Bourdôt et al. 1998).

The second result that highlights the importance of interspecific competition is that C. arvense benefited from competitive release in Nash's Field, showing increased shoot densities on the minus-grass herbicided plots, where perennial grasses like Festuca rubra and Holcus lanatus were at lower abundance following herbicide application. When considering the selective herbicide treatments it is also noteworthy that shoot densities of minus-herb plots remained lower than the controls 3 years after herbicide application. This highlights that selective herbicide application, in combination with the strong competition from grasses that would have occurred, remains a highly effective way of controlling C. arvense.

The third result that highlights the importance of interspecific competition is that N fertilizer and lime application decreased C. arvense shoot density on fenced plots but increased shoot density on grazed plots. On fenced plots, N fertilizer and lime application increased above-ground productivity and standing biomass, particularly of perennial grasses (e.g. Festuca rubra, Holcus lanatus L. and Holcus mollis L.; M.J. Crawley, unpublished data), so enhancing interspecific competition. In contrast, on grazed plots grazing was so concentrated on the N-fertilized and limed plots that, although these plots were more productive, they had similar (or in some years lower) standing biomass than the non-N-fertilized and unlimed plots (Crawley & Rees 1996; M.J. Crawley, unpublished data). The grazed N-fertilized and limed plots for much of the year remained very short, with occasional C. arvense plants, whereas the non-N-fertilized and non-limed plots were tussocky, with conspicuous standing dead matter. The response of grazing to N fertilizer application is similar to that observed in boreal forest herbs (Nams, Folkard & Smith 1996; John & Turkington 1997) and is probably due to the fertilized plots providing a source of more nutrient-rich food for the herbivore.

Effect of phosphorus, magnesium and potassium fertilizers

Phosphorus was the only nutrient other than nitrogen that had any impact on C. arvense, with shoot density increasing in response to P fertilizer application on both grazed and fenced plots. This response may reflect that Nash's Field, with a bicarbonate soil P level at the start of the experiment of < 6 mg kg−1, is considered to be a phosphorus-deficient soil. Future studies must consider specific and threshold effects of phosphorus fertilizer on C. arvense at a range of P fertility levels. This may be particularly relevant given that P fertilizer is regularly applied to boost the productivity of many temperate grasslands. The failure to detect any impact of Mg and K fertilizer may reflect that these nutrients were less limiting in the grassland and so had less impact on biomass of competing species (M.J. Crawley, unpublished data).

Effect of invertebrate herbivores

The impact of insect and mollusc herbivory on C. arvense appears, from our two studies, to be negligible. Recruitment of shoots from fragments and seeds, and cover and shoot density, were similar on pesticide-treated and untreated plots. In studies with other Cirsium species (Louda, Potvin & Collinge 1990; Louda & Potvin 1995) exclusion of capitulum-feeding insects has been shown to increase seedling recruitment and lead to higher mature plant flowering densities in the following generation. In our study, capitulum-feeding insects such as Xyphosia miliaria Schr. and Terellia ruficauda Fabricius were observed in Nash's Field and Oak Mead, but as seedling recruitment appeared to be dependent on the availability of disturbed sites (Heimann & Cussans 1996), rather than the availability of seed, it is possible that these invertebrates may have been consuming seeds that would not have germinated because of a lack of suitable recruitment microsites. The low level of seedling recruitment could also explain why molluscs, which find C. arvense seedlings particularly palatable (Grime, MacPherson-Stewart, & Dearman 1968), failed to have any impact on C. arvense abundance. These concepts could be tested by sowing seeds of C. arvense at a range of seed densities into disturbed and intact grassland with and without invertebrate exclusion treatments.

Management implications: cultivated soil

While repeated cultivation events have been recommended by many authors as a way of controlling C. arvense (reviewed by Donald 1990), single cultivation events in grassland (e.g. those occurring in pasture renovation of grassland restoration programmes) can lead to a dramatic increase in the abundance of the weed. Our study indicates that an effective non-chemical herbicide way of avoiding this potential problem is to sow crops of competing species as soon as possible after cultivation, and delay defoliation (by cutting or grazing) until a ground cover of these species has established. This combination was effective in our study in reducing C. arvense to pre-cultivation levels; seed sowing and delayed grazing will be less effective when imposed as individual treatments. Delayed grazing and seed sowing may be particularly appropriate in the restoration of recreational, conservation or amenity (e.g. roadside) areas, where the need to defoliate recently seeded plants is not as high as it is in pastures (e.g. grass/clover swards). It is likely that seedling crops of most species (e.g. wildflower seeds in conservation areas; forage grasses in arable areas) would exert negative effects on C. arvense, but preference should be given to those that form a closed cover quickly (Hallgren 1976).

Management implications: grassland

Our study indicates that C. arvense abundance in intact grasslands might be reduced by restricting grazing or defoliation of the grassland so that a high biomass of competing vegetation, particularly of perennial grasses, accumulates. This was achieved in our study by no-grazing or single hay cut procedures. Cirsium arvense abundance in intact grassland might also be reduced by the application of fertilizers to stimulate the growth of competing species in combination with grazing regimes that allow the biomass to accumulate. This was achieved in this study by N fertilizer and lime application, but different nutrients may be important in other grasslands, depending on which nutrients are limiting growth. The strong grazing × nutrient (lime) interaction observed in this study indicates that where biomass is grazed or defoliated frequently, fertilizer and lime application may be less effective; in fact they may enhance C. arvense abundance. The formation of bare ground (e.g. that caused by animal digging, treading damage or prolonged overgrazing) should also be avoided as most shoot and seedling recruitment occurs in these areas.

In summary, it is clear that several measures should be considered before implementing any form of control. It is also clear that some procedures may conflict with other management objectives. For instance, while lax or no grazing may reduce C. arvense, it may conflict with objectives such as obtaining maximum grazing for stock, or diversifying the botanical composition of the sward for conservation purposes. However, as C. arvense often occurs in patches, it may be possible to concentrate on these patches, leaving other areas under normal management. Patch treatment may also be appropriate for control measures aimed at regular defoliation of the thistle (e.g. mowing), which rely on destruction of the photosynthetic opportunity of the plant during the growing season and a resultant reduction in root production and succeeding shoot population size (Bourdôt et al. 1998).

Acknowledgements

The research was supported in part by a NERC grant to M.J. Crawley and a New Zealand Science and Technology Post Doctoral Research Fellowship awarded to G.R. Edwards. The authors thank Keith Betteridge, Ian Popay and Mike Hay for their helpful comments on the manuscript.

Received 28 October 1998; revision received 10 December 1999

Ancillary