SEARCH

SEARCH BY CITATION

Keywords:

  • diversification;
  • grazing;
  • hay cutting;
  • Leucanthemum vulgare;
  • Rhinanthus minor

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  • 1
     Diversification of species-poor grassland often requires the introduction of desirable species by sowing seed. Little is known about the factors controlling the spread of introduced species, or how these interact with management. We determined whether management affected spread rates of two grassland species by modifying seed dispersal or seedling establishment.
  • 2
     An experiment was set up in 1995 on a species-poor grassland. It comprised five blocks, each with four treatments: (1) autumn grazed only; (2) cut July; (3) cut July and September; (4) cut July and aftermath grazed. Twenty-two plant species were separately slot-seeded into each treatment plot, providing discrete linear colonization foci.
  • 3
     The mechanisms controlling spread were studied in two species: Rhinanthus minor, an annual with large seeds adapted for wind dispersal; and Leucanthemum vulgare, a perennial with small seeds with no obvious dispersal adaptations.
  • 4
     Perpendicular spread of each species by 1998 was described well by a simple inverse power model. Rhinanthus had spread further in the hay-cut treatments (2–4) than in the grazed treatment (1). Leucanthemum spread poorly in all plots, with no treatment effects.
  • 5
     Seed dispersal from source slots was also described well by the inverse power model. Dispersal curves for Rhinanthus were much longer in the hay-cut treatment (3) than in the grazed treatment (1), because more seed dispersed during hay cutting than before, and cutting dispersed seed longer distances. There was no dispersal by grazing animals. Dispersal showed directional effects: seeds travelled further in the prevailing wind direction before the hay-cut and in the grazed treatment; dispersal by hay cutting was further in the cut direction than in the opposite direction.
  • 6
    Leucanthemum showed poor dispersal, with no treatment effects, except that more seeds were dispersed in the grazed (1) than the hay-cut (3) treatment.
  • 7
     The establishment and survival of sown seeds showed no treatment effects for either species.
  • 8
     Management effects on the spread of Rhinanthus reflected effects on dispersal, rather than establishment. Leucanthemum showed poor dispersal but good establishment in all treatments, suggesting its spread may also have been dispersal-limited. Rhinanthus was positively affected by hay cutting because it set seed at the time of cutting, whereas Leucanthemum set seed later and cutting reduced its seed production.
  • 9
     The results indicate that management of grassland to enhance the colonization of sown species might be best targeted at enhancing seed-dispersal distances. Hay cutting can do this, but must coincide with seed set.

Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

While management is important in restoring the conservation interest of degraded sites, change towards desirable species-rich communities is often hindered by a lack of propagules of new species in the seed bank or seed rain (Bullock et al. 1994; Bakker et al. 1996). Deliberate introduction of new species to a site may form an important – and relatively inexpensive – part of a conservation strategy. There is wide variation in the success of such sown species, in terms of establishment and spread (Pywell et al. 1997), and this is often addressed by sowing large quantities of seed. Stevenson, Bullock & Ward (1995) showed that lower rates than are usually recommended could be used, but that it may take longer to achieve the desired abundance of sown species in the sward. The problem is that little is known about the processes controlling colonization of sown species, or how management may affect this.

The spread of introduced species within a site will be largely controlled by dispersal and establishment (Bakker et al. 1996). These two processes determine the degree to which seeds are spread over a site and whether they can establish as seedlings once they have settled. The relative importance of these two processes is rarely studied, but there is an analogy with seed limitation vs. microsite limitation in established populations. Whether the recruitment of seedlings is limited by lack of seed or by lack of suitable sites for establishment is usually determined by comparing the effects on recruitment of adding seed or disturbing the soil (Crawley 1990; Oesterheld & Sala 1990; Eriksson & Ehrlen 1992). Eriksson & Ehrlen (1992) reported that different studies had shown that both seed and microsite limitation could be important in the regulation of recruitment. They suggested that a combination of both factors might best explain recruitment dynamics.

The same processes have been studied at the landscape scale in terms of the colonization of new sites, asking whether rates of colonization of new sites are limited more by dispersal (the number of seeds reaching a site) or by establishment rates (the number of suitable microsites) (Lee 1993; Scherff, Galen & Stanton 1994; Crawley & Brown 1995). These studies have generally found that dispersal is the limiting factor. In this study we took a similar, but smaller-scale, approach to study the spread of an introduced species within a single site. We asked whether management changes rates of population growth and spread within a site by affecting seed dispersal or the number of microsites. There is much information on how management affects the number of microsites (Silvertown et al. 1992; Bergelson, Newman & Floresroux 1993; Bullock et al. 1994; Cavers, Groves & Kaye 1995; Clear Hill & Silvertown 1997) but little quantitative data on how management affects dispersal. Combine harvesters have been shown to have quantitative effects on weed seed dispersal in arable systems, by lengthening the tail of the dispersal curve (Ballaréet al. 1987; Howard et al. 1991; Ghersa et al. 1993). In grasslands, it has been shown that mowing machinery (Strykstra, Bekker & Verweij 1996; Strykstra, Verweij & Bakker 1997), livestock fleeces (Fischer, Poschlod & Beinlich 1996) and animal guts (Malo & Suarez 1995; Pakeman, Attwood & Engelen 1998) can transport seed both within and between sites, and that species differ in the degree to which they are dispersed by these means. However, there is no information on the changes to grassland species’ dispersal curves under different management systems.

In the study we measured the effects of management (different combinations of grazing and mowing) on dispersal curves and seedling establishment of two species introduced into an experimental diversification of a species-poor grassland. We determined which of these (dispersal or establishment) was the controlling factor in the spread of each species within the grassland, by comparing management effects on establishment and dispersal, respectively, with effects on the spread of populations. Two species with differing seed dispersal characteristics were compared: Rhinanthus minor L., with flattened, winged seeds, and Leucanthemum vulgare Lam., with obovate seeds bearing no obvious specialized structures. Both are found in species-rich mesotrophic grasslands and are commonly introduced in seed mixtures for grassland diversification.

Materials and methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Species description

Rhinanthus minor is a hemi-parasitic summer annual. It is found in a wide range of grassland habitats on soils of low to moderate fertility, but is most typical of hay meadows. The species germinates in spring, requiring chilling to break dormancy. Flowering is from May to August (Grime, Hodgson & Hunt 1988). Seeds are moderately large (3 × 4 mm; van Hulst, Shipley & Theriault 1987) winged discs held loosely in inflated calyces and are set from June onwards. Regeneration is from seed but Rhinanthus has no persistent seed bank, therefore population size is heavily dependent on the production of seed each year. Rhinanthus is capable of limited autotrophic growth (Grime, Hodgson & Hunt 1988) but in the field parasitism is virtually obligate (van Hulst, Shipley & Theriault 1987). However, Rhinanthus is non-host specific and may preferentially parasitize dominant species (Gibson & Watkinson 1989; Davies et al. 1997).

Leucanthemum vulgare is a polycarpic perennial. Achenes are usually shed while inflorescences are dying in August and September (Grime, Hodgson & Hunt 1988), but some seeds are shed from the outer rim of inflorescences earlier in the flowering season (S. Coulson, personal observation). Achenes are 2–3 × 0·8–1 mm (Howarth & Williams 1968) obovate, ribbed, unwinged (Oomes & Elberse 1976) and lack specialized morphological features facilitating dispersal. Regenerative strategies include vegetative lateral spread, although patches are usually less than 50 mm in diameter, and seasonal regeneration by seed (Grime, Hodgson & Hunt 1988). Germination is in late summer, autumn and spring.

Study site

The study took place within a grassland diversification experiment set up in October 1995 at Hill Farm, Little Wittenham, Oxfordshire, UK (15°37′N, 1°10′W). The experimental site was a permanent pasture field on poorly drained Gault Clays of pH 7·4–7·6. The initial vegetation was derived from a resown rye-grass clover ley typical of improved, damp, permanent pasture, dominated by Lolium perenne L., Dactylis glomerata L., Agrostis stolonifera L. and Poa trivialis L. Four treatments were applied; autumn graze only (treatment 1), cut July (treatment 2), cut July and September (treatment 3), cut July and aftermath graze (treatment 4). These reflected hay meadow management options available to land managers. One cut followed by aftermath grazing (treatment 4) is standard hay meadow management. Treatments 2 and 3 comprise cutting only and may be used on farms that do not carry livestock, such as in the east of England where arable farming is predominant and grazing is not an option. This is also true of restoration of grassland on field margins where grazing is not feasible. The second cut (treatment 3) may substitute for an aftermath graze by opening up the sward late in the season to provide establishment microsites. Treatment 1, with only autumn grazing, was carried out to determine whether hay meadow communities can be restored using only grazing. This is an option where farmers can no longer make hay because they have switched to silage production (i.e. they do not have turning machinery or storage facilities). In addition, late grazing is a conservation/restoration management tool specifically designed to facilitate seed dispersal by livestock and to extend the temporal availability of the nectar resource.

Treatments were assigned randomly to four plots in each of five blocks. Each plot measured 20 × 10 m; plots within a block were separated by a 1-m guard row. Blocks were separated by 10-m guard rows. Twenty-eight herb and grass species not found previously in the grassland, including Rhinanthus and Leucanthemum, were slot-seeded into each plot in October 1995, one species per slot. The species were randomly assigned among the slots. A tractor-mounted slot-seeder was used to spray a band of herbicide (approximately 15 cm wide) to kill the existing sward, and to drill the specified species into a slit cut in the ground within the sprayed band. A single pass of the slot-seeder achieved both operations (Wells, Cox & Frost 1989). The slots ran across the 10-m width of each plot, parallel to the plot edge. This design provided relocatable colonization foci for each of the sown species.

Grazing and hay-cut treatments on the plots commenced in 1996. In 1997, a typical year, grazing (treatments 1 and 4) took place between 6 and 31 October at a density of 38·5 ewes ha−1; equivalent to 90·14 livestock units (LSU). In the same year cut plots were cut for hay with an off-line tractor-mounted mower on 16 July (treatments 2–4), the cuttings turned and removed using rakes, and treatment 3 was topped to a height of 10 cm with the same mower on the 19 September.

Seed production: size of seed source for each species

For plots in which seed dispersal was measured, seed heads of Leucanthemum and Rhinanthus plants were counted per metre along the length of the sown slots, across the full width of the band of plants associated with the slots. Twenty-one Rhinanthus plants and 11 Leucanthemum seed heads were removed randomly from the whole experimental area prior to seed set. Seeds per seed head/plant were counted and running means calculated to ensure accurate estimation of mean seed production.

Seed dispersal measurement

Seed dispersal was measured for Rhinanthus and Leucanthemum in two treatments – graze only (treatment 1) and cut July and September (treatment 3) – to provide a contrast between grazing and hay-cut treatments. Measurement sites were selected where there was a clump of plants along a slot that would provide a substantial seed source (at least two flower heads for Leucanthemum and four plants for Rhinanthus) and where there had been minimal spread of plants away from the slot. The latter was more difficult for Rhinanthus, which had spread in some plots. However, all bands of plants associated with the slots were less than 1 m wide, except for that in block 4 treatment 3, which had spread to approximately 2·8 m. Four measurement sites per species were selected for each of the two treatments across the five blocks. Rhinanthus measurement sites were located in blocks 1, 2, 3 and 4 for treatment 3 and blocks 2, 3, 4 and 5 for treatment 1. Leucanthemum dispersal was measured in blocks 1, 2, 3 and 4 for treatment 3 and blocks 2, 3 and 4 for treatment 1; with two measurement sites in block 4 for treatment 1.

At each site seed traps were placed along a transect perpendicular to the slot and crossing the slot through the centre of the clump of target plants. Traps were placed in pairs at the edge of the target clump of plants (0 cm) and at distances of 30, 60, 90, 120, 150, 200, 250, 300, 350 and 400 cm from the clump in both directions. The seed traps were sunk flush with the soil surface and covered with 15-mm gauge wire mesh to protect against damage by livestock and machinery and to reduce seed predation. Traps were put in place 24–26 June 1997, before seeds had begun to be released.

In the hay-cut treatments, traps were checked twice and seeds of the target species removed prior to hay-cut. Following the hay-cut on 16 July 1997, traps were removed (no seed heads remained on the plants), the contents sorted and seeds of the target species counted in the laboratory. In grazed treatments, traps were checked and seeds removed and counted periodically prior to grazing during the period July to October 1997, during grazing 6–31 October and once after grazing ceased on 4 November 1997.

Seed trap data for each species in the hay-cut treatment were split into pre-hay-cut (wind dispersed), and post-hay-cut. Because there was little dispersal in the grazed treatment after grazing started, the data from all trapping dates were merged. Each of these three data sets was then further split into data representing the two directions of trapping away from the source – 61°, roughly east north east (ENE), and 241°, roughly west south west (WSW), for grazed plots, and cut direction/opposite direction for hay-cut plots – giving six data sets for each species (with four replicated lines each). Dispersal curves were modelled by fitting an inverse power relationship relating seed number s to distance d to each of the six data sets:

  • image(eqn 1)

where a indicates the number of seeds falling at the source, and bd indicates the rapidity of the decline in seed numbers with distance from the source. this is the unlogged version of the model used by Willson (1993) and many others to describe a variety of seed shadows. To fit the logged model using linear regression, Willson (1993) had to exclude data points comprising zero values for seed numbers. This is an unnecessary exclusion of valid data values, which gives inaccurate parameter estimates. For this reason the unlogged version was used. However, the variance of the seed number data generally increased with the mean. Therefore the data were square-root transformed and this achieved homogeneity of variances. To simplify interpretation of model parameters the right-hand side of equation 1 was also square-root transformed for model fitting.

Equation 1 was fitted to each data set using PROC NLIN in SAS, which fits models using non-linear least-squares estimation (SAS 1990). The following strategy was used to compare the dispersal curves of two data sets (e.g. cut direction vs. opposite direction), i.e. whether ad1 and ad2 differ, and/or whether bd1 and bd2 differ significantly. Parameters ad1 and bd1 were estimated for the first data set. Then addiff and bddiff were estimated, which represented the difference between ad1 and ad2, bd1 and bd2, respectively. The 95% confidence intervals for, ad1, bd1, addiff and bddiff were then used to determine whether the ad and bd estimates differed significantly between the data sets.

The same procedure was used to compare dispersal curves between the two species. Data sets for the two species from the same treatment and period were compared (e.g. pre-hay-cut ENE). An exception was the post-hay-cut, opposite direction data sets, where there was no model convergence for the Leucanthemum data set, so the Leucanthemum post-hay-cut, cut direction data set was compared with the Rhinanthus post-hay-cut, opposite direction data set.

Thirty samples of sheep dung were collected randomly from the entire grazed area of the experiment. These were washed through a 1-mm gauge sieve to separate out any Rhinanthus or Leucanthemum seeds that had passed through the sheep.

Seedling establishment and survival

Seedling establishment was measured for both species, in each of the four treatments, in blocks 3–5 (three out of five blocks were used due to time constraints). A transect running parallel to the seed slots was placed randomly, within the central 6 m of each plot, but avoiding sown slots and, where possible, spreading Rhinanthus. Along the transect, a permanent 50 × 30-cm quadrat was placed every metre, excluding the top and bottom 2 m of each plot and the guard rows, giving seven quadrats per plot. Of these seven quadrats, three were assigned randomly to each species (Rhinanthus and Leucanthemum) and one to an unsown control. Seeds were sown individually using forceps, at 5-cm intervals, forming grids of 10 × 6 seeds in each permanent quadrat. Sowing took place over 2 days, 23–24 August 1997.

Half of the quadrats were checked for seedlings after approximately 3 weeks, the second half after approximately 4 weeks. Thereafter seedling establishment was recorded for each set of quadrats at approximately 3-week intervals, over a period of 9·5 months. Recording of seedlings was restricted to those occurring within 1 cm of the grid position in the quadrat. In control quadrats, establishment was recorded for a grid corresponding to those used in sown quadrats.

Spread measurement

The spread of Leucanthemum and Rhinanthus away from the sown slots was measured on 15 June 1998 by laying three transects at random positions across each plot (4 treatments × 5 blocks) perpendicular to and crossing the sown slots. Transects consisted of lines of 1 × 0·5-m quadrats. Percentage cover was estimated for both species in each quadrat, giving data describing the variation in cover of each species across each plot.

Examination of the spread data indicated a negative relationship with distance. Therefore the parameters describing the rate of spread away from the slot were estimated using the same model as for the seed dispersal curves to relate percentage cover c to distance d.

  • image(eqn 2)

where as is cover at 0[middot]5[nbsp]m from the slot (see below) and bs is the rapidity of the decline in cover with distance d. percentage cover at each distance was calculated as the mean of values from the three transects within each plot, and was arcsine transformed to achieve normality. Equation 2 was fitted separately for each treatment in each direction, with the blocks as replicates, using the PROC NLIN procedure. Cover values at 0 m, the slot position, were excluded from the analyses, because these did not represent spread away from the slot. The two directions within each treatment were compared using the same methods as described for comparing dispersal curves. Only the Rhinanthus treatment 1 spread data showed significant effects of direction on parameter estimates (see below), so the data for the two directions in each treatment were merged and models fitted to each treatment. The parameter estimates for each of the four treatments were then compared in all pair-wise combinations. These comparisons were carried out only on the Rhinanthus data. Leucanthemum showed very little spread beyond 0·5 m from the slot in any treatment (see below), so comparison of model parameters was irrelevant. The same procedure was used to compare spread of the two species within each treatment.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Seed production: size of seed source for each species

Leucanthemum had a mean of 250·8 seeds head−1 (SE = 27·2, n = 11) and Rhinanthus had a mean of 92·9 seeds plant−1 (SE = 11·3, n = 21). Combining the head/plant count for the metre containing the trap position (Leucanthemum = 10·1 ± 2·1, Rhinanthus = 49·4 ± 15·8) with the mean seeds per head/plant, describes the seed source for the trapping position: Rhinanthus = 4589 seeds m−1, Leucanthemum = 2543 seeds m−1. anova showed that for Rhinanthus there were no differences between treatments 1 and 3 in the numbers of plants at the trap positions (F1,2 = 1·73, NS) nor along the whole slots (F1,2 = 0·07, NS). Leucanthemum also showed no differences in source size at either the trap positions (F1,2 = 12·52, NS) or along the whole slots (F1,2 = 0·02, NS). Therefore treatment effects on dispersal curves were not caused by differences in source size.

Seed dispersal

Good model fits were achieved for all dispersal data sets (Figs 1 and 2) except for one in which only one seed was trapped (Fig. 1d). The percentage of variation explained was between 37% and 83%. Leucanthemum seeds dispersed poorly in all treatments, with few seeds travelling further than 0·6 m. As a consequence, all Leucanthemum data sets had a low bd for equation 1 (Fig. 1) and there were few differences between treatments or directions in the ‘dispersal distance’, i.e. the rate of decline in seed numbers with distance (bd) (Table 1). However, within the hay-cut treatment there was a directional effect, in that the dispersal distance was greater towards the ENE rather than the WSW. There were more effects on ad, the number of seeds falling at the source and an indicator of the total number of seeds dispersed. Within the hay-cut treatment, almost all seeds were dispersed before the hay-cut, with hardly any trapped after the cut. Significantly more seeds were dispersed in the grazed treatment than in the hay-cut – probably because cutting removed unripe seed and seed heads – but there were no differences in dispersal distance. The amount of seed dispersed was greater in the ENE direction than the WSW direction in both the grazed and pre-cut data sets.

image

Figure 1. Parameter values estimated for each of the Leucanthemum seed dispersal data sets according to the model seed number =ad · bdd. The points give the measured seed numbers at distances from the source slot for each of the four replicate trap lines, and the curve is the fitted model. The model was fitted with both sides of the equation square root transformed, but the back-transformed data and model estimates are shown here. The parameter estimation did not reach convergence for the post-hay-cut, opposite direction data. n was 44 for each data set. Note the y-axis scales are different for each graph. *P < 0·05, **P,0·01, ***P < 0·001.

Download figure to PowerPoint

image

Figure 2. Parameter values estimated for each of the Rhinanthus seed dispersal data sets according to the model seed number =ad·bdd. The points give the measured seed number at distances from the source slot for each of the four replicate trap lines, and the curve is the fitted model. n was 44 for each data set. Note the y-axis scales are different for each graph.

Download figure to PowerPoint

Table 1.  Comparisons between the parameter estimates of selected pairs of seed dispersal data sets for Leucanthemum. The label for each data set is that given in Fig. 1 and the labels in the a and b columns give the data set with the higher parameter value of the pair. – indicates no significant difference. Significance is at the 5% level (see text)
Data sets comparedab
(a) Pre-cut east vs. (b) pre-cut westaa
(a) Pre-cut east vs. (c) post-cut cut directionaa
(b) Pre-cut west vs. (c) post-cut cut direction
(a) Pre-cut east vs. (e) grazed easte– 
(a) Pre-cut east vs. (f) grazed west
(e) Grazed east vs. (f) grazed weste

In contrast, most seed dispersal of Rhinanthus in the hay-cut treatment occurred after the hay-cut (Fig. 2), so both the amount of seed and distance dispersed were significantly greater in the post-cut data sets than in the pre-cut (Table 2). Post-cut, there were greater dispersal distances and more seeds dispersed in the direction of the hay-cut than in the opposite direction. Hay cutting, in both cut and opposite directions, was also more effective in dispersing seeds than grazing, in terms of both numbers of seeds and distance dispersed. In both pre-cut and grazing data sets seeds were dispersed further in the ENE direction than the WSW, and in the pre-cut data more seeds were also dispersed ENE.

Table 2.  Comparisons between the parameter estimates of selected pairs of seed dispersal data sets for Rhinanthus. The label for each data set is that given in Fig. 2 and the labels in the a and b columns give the data set with the higher parameter value of the pair. – indicates no significant difference. Significance is at the 5% level (see text)
Data sets comparedab
(a) Pre-cut east vs. (b) pre-cut westaa
(a) Pre-cut east vs. (c) post-cut cut directioncc
(a) Pre-cut east vs. (d) post-cut opposite directiond
(c) Post-cut cut direction vs. (d) post-cut opposite directioncc
(c) Post-cut cut direction vs. (e) grazed eastcc
(d) Post-cut opposite direction vs. (e) grazed eastd
(e) Grazed east vs. (f) grazed weste

The similar directional effects on dispersal in the pre-cut and grazed data sets related to directional differences in wind speed. In the grazed treatment most seeds were dispersed before grazing began (Leucanthemum= 94·9%, Rhinanthus= 94·4%) so, as for the pre-cut data sets, most seeds must have been dispersed by wind. The fact that no seeds of either species were found in the dung samples provided supplementary evidence for this conclusion.

Daily wind speed and direction at midday were obtained from the Benson meteorological station (51°62′N, 1°10′W, 7 km from the experimental site) for the duration of dispersal data collection. The ‘cumulative wind speed’ was calculated for each direction (in 10° intervals), this being the sum of wind speeds over all measurement times the wind was in that direction. This described both wind speed and the proportion of time it was in a particular direction. Seeds would have been blown along the ENE transects by winds from directions 160°–330° and along the WSW transects by winds from 340° to 150°. Cumulative wind speeds were greater for the ENE transects (mean = 9·6 days m s−1) than the WSW transects (5·3 days m s−1) (Fig. 3).

image

Figure 3. The cumulative wind speed days m−1 s−1 (the sum of wind speeds over all daily measurement points at which the wind was blowing from that direction) for each compass direction between 1 June and 30 September 1997.

Download figure to PowerPoint

The above treatment and directional effects on each species were reflected in the differences in dispersal curves between the two species. Rhinanthus dispersed more seed and further than Leucanthemum in the hay-cut treatment (i.e. post-hay-cut), but there were no consistent differences between species in the grazed treatment (Table 3).

Table 3.  Comparisons between the parameter estimates of Rhinanthus and Leucanthemum for each pair of seed dispersal data sets. The species in the a and b columns is the one with the higher parameter value of the pair. – indicates no significant difference. Significance is at the 5% level (see text). For Leucanthemum the post-hay-cut, cut direction data set is compared with the Rhinanthus post-hay-cut, opposite direction data set because there was no model convergence for the Leucanthemum post-hay-cut, opposite direction data set
Data setab
Pre-cut east
Pre-cut west
Post-cut cut directionRhinanthusRhinanthus
Post-cut opposite directionRhinanthusRhinanthus
Grazed eastLeucanthemum
Grazed westLeucanthemum

Establishment

Rhinanthus seedlings were found in both the control and Leucanthemum quadrats, so these data were combined to give percentage establishment at grid positions in ‘unsown’ quadrats. Mean percentage establishment in ‘sown’ and unsown quadrats was calculated for each plot and subjected to split-plot anova (sowing as split-plot factor, treatment as main factor) using arcsine-transformed data. Sowing had a significant effect on percentage establishment (F1,8 = 9·32, P < 0·05), with sown plots showing greater establishment than unsown (Fig. 4). Treatment effects were also significant (F3,6 = 10·29, P < 0·01), with differences in establishment of the order 4 > 2,1 for unsown quadrats and of the order 4 > 2,1 and 3 > 1 for sown quadrats (Tukey multiple comparison for family error rate = 0·05). However, there was no treatment–sowing interaction (F3,8 = 2·42, NS), indicating there were similar treatment effects in both sown and unsown quadrats. To remove the effect of background germination, the difference between unsown and sown quadrats was calculated for each plot. Due to large variation within treatments, anova showed no significant treatment effect on this difference (F3,6 = 2·57, NS), although there was a trend 4 > 3 > 2 > 1 (Fig. 4).

image

Figure 4. Proportionate establishment (with standard errors) of Rhinanthus minor in sown and unsown quadrats and for sown quadrats with the effects of background germination accounted for (sown–unsown). Treatments compared were (1) autumn grazed; (2) hay-cut in July; (3) hay-cuts in July and September; (4) hay-cut in July with aftermath grazing.

Download figure to PowerPoint

There was no background germination of Leucanthemum and this species showed no treatment effect on establishment in sown quadrats (mean proportion establishing = 0·057, F3,8 = 1·66, NS), but in the sown quadrats of each species establishment was greater for Leucanthemum than for Rhinanthus (Rhinanthus data were corrected for background germination; split-plot anova, F1,8 = 12·13, P < 0·01).

Survival

Survival of germinated Leucanthemum seedlings was higher than for Rhinanthus (sown quadrats only) seedlings (F1,5 = 7·81, P < 0·05). There was no treatment effect on seedling survival for either species (Leucanthemum mean = 76·4%, F3,5 = 3·91, NS; Rhinanthus mean = 44·2%, F3,7 = 2·62, NS).

Spread

Model fits were good for all data sets (Figs 5 and 6) with r2 values between 44% and 82%. For Rhinanthus, percentage cover near the slot (comparing as) was not affected by treatment, but hay-cutting (treatments 4, 3 and 2) allowed further spread than grazing only (treatment 1), and hay-cutting with grazing (treatment 4) showed further spread than cutting alone (treatment 2) (comparing bs) (Fig. 5 and Table 4).

image

Figure 5. Parameter values estimated for the spread from sown slots of Rhinanthus in each treatment according to the model c = as·bsd. Treatments were (1) autumn grazed; (2) hay-cut in July; (3) hay-cuts in July and September; (4) hay-cut in July with aftermath grazing. The points give the measured cover for each of the five blocks in both directions from the slot, and the curve is the fitted model. The percentage cover is arcsine transformed (see text).

Download figure to PowerPoint

image

Figure 6. Parameter values estimated for the spread from sown slots of Leucanthemum in each treatment according to the model c = as· bsd. Treatments were (1) autumn grazed; (2) hay-cut in July; (3) hay-cuts in July and September; (4) hay-cut in July with aftermath grazing. The points give the measured cover for each of the five blocks in both directions from the slot, and the curve is the fitted model. The percentage cover is arcsine transformed (see text).

Download figure to PowerPoint

In treatment 1 spread was further in the ENE direction than in the WSW (Fig. 7), mirroring differences in cumulative wind speed, but there were no directional effects in the hay-cut treatments (treatments 2, 3 and 4).

image

Figure 7. Parameter values estimated for the spread from sown slots of Rhinanthus to the ENE and WSW from sown slots in treatment 1 according to the model c = as· bsd. Treatments were (1) autumn grazed; (2) hay-cut in July; (3) hay-cuts in July and September; (4) hay-cut in July with aftermath grazing. The points give the measured cover for each of the five blocks, and the curves represent the fitted model. The percentage cover is arcsine transformed (see text).

Download figure to PowerPoint

Leucanthemum showed negligible spread beyond the first 0·5 m in all treatments and there were no treatment effects on spread (Fig. 6). Therefore Rhinanthus showed greater spread than Leucanthemum in all the hay-cut treatments (2–4), but there were no species effects in the grazed treatment 1 (Table 5).

Table 5.  Comparisons between the parameter estimates of Rhinanthus and Leucanthemum for spread in each treatment. The species in the a and b columns is the one with the higher parameter value of the pair. – indicates no significant difference. Significance is at the 5% level (see text)
Treatmentab
Treatment 1
Treatment 2Rhinanthus
Treatment 3Rhinanthus
Treatment 4Rhinanthus

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Dispersal vs. microsite limitation in the spread of colonising species

Rate-limiting processes in the spread of introduced species within sites have not been studied in this way before. Studies of microsite vs. seed limitation in established populations (Oesterheld & Sala 1990) have considered these processes, but not in a spatially explicit manner. Larger scale studies of the ability of plants to colonize new sites in a landscape have shown that a spatially explicit approach is important (Lee 1993; Scherff, Galen & Stanton 1994; Crawley & Brown 1995). For example, Crawley & Brown (1995) found that the establishment of new Brassica rapa oleifera populations along both verges of the M25 motorway was seed-limited. While colonization foci in our study were highly artificial straight lines, the same concepts and processes are important whether colonization is by natural means or by other management activities, such as broadcast sowing or transport on machinery. In all cases the initial population is very patchy and must spread out from these foci (Cousens & Mortimer 1995; Verdu & Garcia Fayos 1998). Therefore there is a need for more work on the relative contributions of dispersal and establishment limitation to controlling rates of increase in colonizing populations.

The patterns of treatment effects on spread of Rhinanthus mirror effects on dispersal rather than on establishment or survival. There was greater dispersal in hay-cut treatment 3 than in grazed treatment 1, while hay-cut treatments 4, 3 and 2 all showed greater spread than treatment 1. Establishment and survival, once corrected for higher background seed densities in the hay-cut treatments, showed no or very weak responses to treatment (however, the fact that 4 showed greater spread than 2 may reflect weak effects on establishment). The directional effects on spread observed in treatment 1 reflect identical effects on dispersal, which were caused by stronger and more frequent winds in the ENE direction. Lack of any directional effect in treatments 2, 3 and 4 reflects the fact that cutting was not in one uniform direction within plots or between years. For Leucanthemum, the poor spread overall and the lack of treatment effects reflect the lack of treatment effects on dispersal, establishment or survival. There is some evidence that spread was dispersal-limited because spread was much less than that found for Rhinanthus, as was dispersal, while establishment and survival were higher for Leucanthemum. However, differences in spread may reflect other species differences, such as the perennial vs. annual life history, so this conclusion is tenuous.

Dispersal

The effect of wind direction on dispersal of both species might be expected, but is rarely reported (Carey & Watkinson 1993). This could be important in patterns of colonization, for example Ash, Gemmel & Bradshaw (1994), in an experimental introduction by seeding onto a blast furnace slag waste heap, found that, in the absence of human interference, the Rhinanthus population had spread rapidly to the limits of suitable habitat in the downwind direction. This is despite the assertion by van Hulst, Shipley & Theriault (1987) that the seeds of Rhinanthus minor are too big to be readily dispersed under natural conditions.

Cutting did not increase dispersal distances in Leucanthemum compared with the grazed or pre-cut data sets. In fact, cutting appeared to hinder dispersal, probably because the early (July) cutting removed heads containing unripe seeds that were consequently not dispersed within the experimental area. The majority of L. vulgare seed is shed in August and September (Howarth & Williams 1968). Prior to this, only small amounts of seed are shed from around the periphery of the inflorescence (S. Coulson, personal observation).

In contrast to the lack of effect on Leucanthemum, cutting had a major effect on the dispersal of Rhinanthus. ter Borg (1983) reported an unpublished study that found that mowing during seed ripening resulted in dispersal of over 2 m for Rhinanthus serotinus and R. minor, and that the hay-turning machine added another 6–7 m to the maximum dispersal distance. In our study, it is unlikely that hand-turning of hay with rakes would increase dispersal distance.Howard et al. (1991) found that seed dispersal from Bromus species (in arable weed populations) followed simple Gaussian distributions, centred on the source plant, in the absence of human interference. However, combine harvesting skewed the distribution of seeds, such that there was further spread in the movement direction of the harvester. Similarly Ghersa et al. (1993) found that combine harvesting spread Sorghum halepense seeds much further (50 m in the direction of harvesting) compared with wind dispersal (6·3 m). This reflects our findings for Rhinanthus dispersal, where hay cutting moved seeds up to at least 4 m in the cut direction, compared with a maximum 0·9 m in the grazed treatment. Somewhat unexpectedly, maximum dispersal distances were increased in the direction opposite to the hay-cut as well (2 m). Strykstra, Verweij & Bakker (1997) found that different parts of a skid disk mower moved seeds in different ways: material accumulated on the safety skirt and fell off when the mowing machinery was raised or when the mower hit irregularities in the field; material on the skid disk remained in situ until wiped off by other vegetation. In our study, the smoothness of the tail of the dispersal curve and the two-directional effect of mowing on dispersal suggested that dispersal was increased by the action of the cutting blades, rather than by seed sticking to the machinery and dropping later on. The latter would only move seed in the direction in which the mower was moving and would be more likely to produce irregularities in the tail of the dispersal curve.

Grazing had little effect on dispersal of either species, because grazing started after most of the seeds had been dispersed. The remaining ripe seed was unlikely to have been dispersed in the guts of the grazing animals (endozoochory), as this seed would have been on plants whose stems and inflorescences would be dead or dying, and hence unpalatable, by the time grazing commenced. Epizoochory of seeds on animal fleeces or hoofs is possible, but neither species have specialized structures to facilitate this. Fischer, Poschlod & Beinlich (1996) found very few Leucanthemum seeds were transported on sheep.

Establishment and survival

Much of the variation in establishment between treatments for Rhinanthus was due to different numbers of unsown seeds in the background. This provides further evidence for the differential effects of treatment on dispersal, with hay-cut plots showing a larger amount of background establishment than grazed plots. There was still a non-significant trend for treatment effects once background establishment was accounted for, but considering that 180 seeds were sown in each plot and 540 seeds in each treatment, giving a large sample size, any treatment effect on establishment was weak. Establishment and survival of forb seedlings in grasslands has often been shown to be increased by grazing (Silvertown et al. 1992; Bullock et al. 1994; Clear Hill & Silvertown 1997) or hay-cutting (Hutchings & Booth 1996; Spackova, Kotorova & Leps 1998). These management operations create gaps in which seedlings can escape competition from established plants (Bullock et al. 1995). However, the effects of cutting vs. grazing are not usually compared (but see Bakker & de Vries 1992) and in this case they showed little (Rhinanthus) or no (Leucanthemum) difference in their effect on seedling establishment or survival.

Applications

This study suggests that management to facilitate or increase rates of spread of introduced species should be targeted at increasing dispersal distances. This is particularly important when slot-seeding is used, as plants are introduced in an extremely spatially limited manner. However, slot-seeding is more effective in successfully establishing plants in existing grassland than other methods such as broadcast sowing (Wells, Cox & Frost 1989). If management such as mowing can be used to increase dispersal distances, it is clear that the timing of mowing is critical. Cutting date is indeed important for species composition and diversity of grasslands (Kirkham & Tallowin 1995; Smith et al. 1996a,b). Smith, Pullan & Shiel (1996c) found that the number of seeds reaching the ground in a mesotrophic grassland was affected by cutting date for a number of species. Cut dates were 14 June, 21 July or 3 September, and a later cut date increased the crop for many species. Rhinanthus minor was one species for which the cut date was not important.

In our study the cutting date of mid-July seemed to be ideal for seed dispersal and spread of Rhinanthus, which sets ripe seed from June–July (Grime, Hodgson & Hunt 1988). However, Leucanthemum sets seed later, mostly in August (Grime, Hodgson & Hunt 1988), and the cut date was too early to allow much seed to ripen or to facilitate dispersal. Of the remaining 20 forb species sown in this restoration experiment (R.F. Pywell, unpublished data), the earliest month in which ripe seed appears is June for five (Cerastium fontanum, Plantago lanceolata, Ranunculus bulbosus, Rumex acetosa, Trifolium dubium), July for nine (Achilleamillefolium, Centaureanigra, Leontodonhispidus, Lotuscorniculatus, Plantagomedia, Primulaveris, Ranunculusacris, Sanguisorbaminor, Trifoliumpratense), August for five (Agrimonia eupatoria, Knautia arvensis, Lathyrus pratensis, Prunella vulgaris, Vicia cracca) and September for Galium verum (Grime, Hodgson & Hunt 1988).

Many grasslands in Britain are managed under environmentally sensitive area (ESA) agreements, by which government-funded incentives are offered to farmers to adopt agricultural practices that will safeguard and enhance the rural environment. In the past there has been some concern that cutting dates prescribed for ESA in Britain were generally earlier than would have been used traditionally, and may have been too early to maintain species diversity (Smith & Jones 1991). The recent revision of management prescriptions for the Upper Thames Tributaries ESA (near our experimental site) requires cutting after 1 July for Tiers 1B (extensive grassland management) and 3A (arable reversion to species-rich grassland). Cutting has to be after 10 July for the ‘traditional hay-making supplement’ in these tiers. From the seed-set dates given above it seems that a mid-July cutting date would be beneficial for only the six June species. The others would have produced little or no seed by this cut date – even those with the first ripe seed appearing in July would have produced little by this date – and would probably lose most of their potential seed crop by the removal of flowers. Many (15) of the 22 sown forb species in our experiment can reproduce vegetatively (Grime, Hodgson & Hunt 1988), but they could only maintain local patches by this process in the absence of seed set. Their ability to spread rapidly from introduction foci would still be governed by seed dispersal. Some species are able to flower again and set seed after an early hay-cut (e.g. Lathyrus pratensis, Lotus corniculatus), so there may not be a complete loss of the seed crop. However, the potential for increased seed dispersal by hay-cutting would then be lost. Therefore despite the ability of some species to reproduce vegetatively or to reshoot after cutting, cut date may nevertheless be even more critical in determining rates of spread of introduced species than it is for the maintenance of populations of established species.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

We thank Kevin Walker, Bill Meek, Dave Barratt, Neil Berril, Steve Gregory, Susanna Kay and Rachel Hayward who kindly helped with monitoring, and Liz Allchin who sorted the dung samples. Our thanks also to John Sargent, Rob Dingle and the Northmoor Trust Estate team for maintaining the experiment and Ralph Clarke for giving valuable statistical advice. Two referees gave useful comments on an earlier draft.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  • Ash, H.J., Gemmel, R.P., Bradshaw, A.D. (1994) The introduction of native plant species on industrial waste heaps – a test of immigration and other factors affecting primary succession. Journal of Applied Ecology, 31, 7484.
  • Bakker, J.P. & De Vries, Y. (1992) Germination and early establishment of lower salt-marsh species in grazed and mown salt marsh. Journal of Vegetation Science, 3, 247252.
  • Bakker, J.P., Poschlod, P., Strykstra, R.J., Bekker, R.M., Thompson, K. (1996) Seed banks and seed dispersal: important topics in restoration ecology. Acta Botanica Neerlandica, 45, 461490.
  • Ballare, C.L., Sopel, A.L., Ghersa, C.M., Sanchez, R.A. (1987) The population ecology of Datura ferox in soybean crops – a simulation approach incorporating seed dispersal. Agriculture Ecosystems and Environment, 19, 177188.
  • Bergelson, J., Newman, J.A., Floresroux, E.M. (1993) Rate of weed spread in spatially heterogeneous environments. Ecology, 74, 9991011.
  • Ter Borg, S.J. (1983) Population biology and habitat relations of some hemiparasitic Scrophulariaceae. The Population Structure of Vegetation (ed. J.White), pp. 463487. Junk, Dordrecht, the Netherlands.
  • Bullock, J.M., Clear Hill, B., Dale, M.P., Silvertown, J. (1994) An experimental study of vegetation change due to sheep grazing in a species-poor grassland and the role of seedling recruitment into gaps. Journal of Applied Ecology, 31, 493507.
  • Bullock, J.M., Clear Hill, B., Silvertown, J., Sutton, M. (1995) Gap colonization as a source of grassland community change: effects of gap size and grazing on the rate and mode of colonisation by different species. Oikos, 72, 273282.
  • Carey, P.D. & Watkinson, A.R. (1993) The dispersal and fates of seeds of the winter annual grass Vulpia ciliata. Journal of Ecology, 81, 759767.
  • Cavers, P.B., Groves, R.H., Kaye, P.E. (1995) Seed population dynamics of Onopordum over 1 year in southern New South Wales. Journal of Applied Ecology, 32, 425433.
  • Clear Hill, B.H. & Silvertown, J. (1997) Higher-order interaction between molluscs and sheep affecting seedling numbers in grassland. Acta Oecologica – International Journal of Ecology, 18, 587596.
  • Cousens, R. & Mortimer, M. (1995) Dynamics of Weed Populations. Cambridge University Press, Cambridge, UK.
  • Crawley, M.J. (1990) The population dynamics of plants. Philosophical Transactions of the Royal Society of London Series B, Biological Sciences, 330, 125140.
  • Crawley, M.J. & Brown, S.L. (1995) Seed limitation and the dynamics of feral oilseed rape. Proceedings of the Royal Society of London B, 259, 4954.
  • Davies, D.M., Graves, J.D., Elias, C.O., Williams, P.J. (1997) The impact of Rhinanthus spp. on sward productivity and composition: implications for the restoration of species-rich grasslands. Biological Conservation, 82, 8793.
  • Eriksson, O. & Ehrlen, J. (1992) Seed and microsite limitation of recruitment in plant populations. Oecologia, 91, 360364.
  • Fischer, S.F., Poschlod, P., Beinlich, B. (1996) Experimental studies on the dispersal of plants and animals on sheep in calcareous grasslands. Journal of Applied Ecology, 33, 12061222.
  • Ghersa, C.M., Ghersa-Martinez, M.A., Sattore, E.H., Van Esso, M.L., Chicotky, G. (1993) Seed dispersal distribution and recruitment of seedlings of Sorghum halopense L. Weed Research, 33, 7988.
  • Gibson, C.C. & Watkinson, A.R. (1989) The host range and selectivity of a parasitic plant –Rhinanthus minor L. Oecologia, 78, 401406.
  • Grime, J.P., Hodgson, J.G., Hunt, R. (1988) Comparative Plant Ecology. Unwin Hyman, London, UK.
  • Howard, C.L., Mortimer, A.M., Gould, A.M., Putwain, P., Cousens, R., Cussans, G.W. (1991) The dispersal of weeds: seed movement in agriculture. Proceedings of the Brighton Crop Protection Conference – Weeds, pp. 505514. British Crop Protection Council, Farnham, UK.
  • Howarth, S.E. & Williams, J.T. (1968) Chrysanthemum leucanthemum. Biological flora of the British Isles. Journal of Ecology, 56, 585595.
  • Van Hulst, R., Shipley, B., Theriault, A. (1987) Why is Rhinanthus minor (Scrophulariaceae) such a good invader? Canadian Journal of Botany, 65, 23732379.
  • Hutchings, M.J. & Booth, K.D. (1996) Studies on the feasibility of recreating chalk grassland vegetation on ex-arable land. II. Germination and early survivorship of seedlings. Journal of Applied Ecology, 33, 11821190.
  • Kirkham, F.W. & Tallowin, J.R.B. (1995) The influence of cutting date and previous fertiliser treatment on the productivity and botanical composition of species-rich hay meadows on the Somerset Levels. Grass and Forage Science, 50, 365377.
  • Lee, P.C. (1993) The effect of seed dispersal limitation on the spatial distribution of a gap species, seaside goldenrod (Solidago sempervirens). Canadian Journal of Botany, 71, 978984.
  • Malo, J.E. & Suarez, F. (1995) Cattle dung and the fate of Biserrula pelecinus L. (Leguminosae) in a Mediterranean pasture – seed dispersal, germination and recruitment. Botanical Journal of the Linnean Society, 118, 139148.
  • Oesterheld, M. & Sala, O.E. (1990) Effects of grazing on seedling establishment: the role of seed and safe-site availability. Journal of Vegetation Science, 1, 353358.
  • Oomes, M.J.M. & Elberse, W.T. (1976) Germination of six grassland herbs in microsites with different water contents. Journal of Ecology, 64, 745755.
  • Pakeman, R.J., Attwood, J.P., Engelen, J. (1998) Sources of plant colonising experimentally disturbed patches in an acidic grassland, eastern England. Journal of Ecology, 86, 10321040.
  • Pywell, R.F., Peel, S., Hopkins, A., Bullock, J.M. (1997) Multi-site experiments on the restoration of botanically diverse grasslands in ESAs. Grassland Management in Environmentally Sensitive Areas. BGS Occasional Symposium 32 (ed. R.D.Sheldrick), pp. 160167. British Grassland Society, Reading, UK.
  • SAS (1990) SAS/STAT Users’ Guide. SAS Institute, Carey, NC.
  • Scherff, E.J., Galen, C., Stanton, M.L. (1994) Seed dispersal, seedling survival and habitat affinity in a snowbed plant – limits to the distribution of the snow buttercup, Ranunculus adoneus. Oikos, 69, 405413.
  • Silvertown, J., Watt, T.A., Smith, B., Treweek, J.R. (1992) Complex effects of grazing treatment on the establishment, growth, survival and reproduction of an annual in a species-poor grassland community. Journal of Vegetation Science, 3, 3540.
  • Smith, R.S. & Jones, L. (1991) The phenology of mesotrophic grassland in the Pennine dales, Northern England: historic hay cutting dates, vegetation variation and plant species phenologies. Journal of Applied Ecology, 28, 4259.
  • Smith, R.S., Buckingham, H., Bullard, M.J.S., Shiel, R.S., Younger, A. (1996a) The conservation management of mesotrophic (meadow) grassland in Northern England. I. Effects of grazing, cutting date, fertiliser and seed application on the vegetation of an agriculturally improved sward. Grass and Forage Science, 51, 278291.
  • Smith, R.S., Corkhill, P., Shiel, R.S., Milward, D. (1996b) The conservation management of mesotrophic (meadow) grassland in Northern England. II. Effects of grazing, cutting date, fertilizer and seed application on the vegetation of an agriculturally improved sward. Grass and Forage Science, 51, 292305.
  • Smith, R.S., Pullan, S., Shiel, R.S. (1996c) Seed shed in the making of hay from mesotrophic grassland in a field in northern England: effects of hay cut date, grazing and fertilizer in a split-split-plot experiment. Journal of Applied Ecology, 33, 833841.
  • Spackova, I., Kotorova, I., Leps, J. (1998) Sensitivity of seedling recruitment to moss, litter and dominant removal in an oligotrophic wet meadow. Folia Geobotanica, 33, 1730.
  • Stevenson, M.J., Bullock, J.M., Ward, L.K. (1995) Recreating semi-natural communities: effect of sowing rate on establishment of calcareous grassland. Restoration Ecology, 3, 279289.
  • Strykstra, R.J., Bekker, R.M., Verweij, G.L. (1996) Establishment of Rhinanthus angustifolius in a successional hayfield after seed dispersal by mowing machinery. Acta Botanica Neerlandica, 45, 557562.
  • Strykstra, R.J., Verweij, G.L., Bakker, J.P. (1997) Seed dispersal by mowing machinery in a Dutch brook valley system. Acta Botanica Neerlandica, 46, 387401.
  • Verdu, M. & Garcia Fayos, P. (1998) Old-field colonisation by Daphne gnidium: seedling distribution and spatial dependence at different scales. Journal of Vegetation Science, 9, 713718.
  • Wells, T.C.E., Cox, R., Frost, A. (1989) Diversifying grasslands by introducing seed and transplants into existing vegetation. Biological Habitat Reconstruction (ed. G.P.Buckley), pp. 283298. Bellhaven Press, London, UK.
  • Willson, M.F. (1993) Dispersal mode, seed shadows, and colonisation patterns. Vegetatio, 108, 261280.

Received 5 August 1999; revision received 8 August 2000