M.G. Le Duc, Applied Vegetation Dynamics Laboratory, School of Biological Sciences, University of Liverpool, PO Box 147, Liverpool L69 3BX, UK (fax +44 151 7944940; e-mail firstname.lastname@example.org).
1Bracken Pteridium aquilinum is a serious weed of upland and marginal land. Its extensive rhizome system and large carbohydrate reserves make control difficult. This paper reports the results of seven long-term experiments, established in four diverse UK locations, to test control and vegetation restoration treatments.
2Samples were obtained from 580 rhizome pits between 1998 and 2000. Total dry mass per unit area (M, a measure of performance) and ratio of frond-bearing to total rhizome dry mass (R, a relative measure of investment in frond production) were measured.
3The range of means for M in untreated bracken was 1·8–5·1 kg m−2, greater than that reported by others (1·2–3·0 kg m−2). Measured values for R were 0·24–0·42, again differing from other reports (0·10–0·38), probably due to our policy of avoiding advancing bracken fronts.
4Randomization tests were used to check for non-independence of sampling units. They confirmed the general adequacy of the anova results; there was little detectable interference between plots. Two regions contained matching pairs of experiments, thus variation within region was tested and differences were found in one region.
5Five bracken control treatments (cutting, herbicide application and combinations) were employed at all but one experiment. Cutting once or twice per year generally gave the best results, reducing M by c. 60% after ≤ 5 years. Some subtreatments designed mainly for vegetation restoration, notably surface disturbance, also affected M.
6R was reduced by herbicide treatment, for example from 0·30 to 0·16. Such low values of R are typical of invading bracken and are thought to represent rejuvenated and invasive rhizomes. Follow-up treatments are needed in such situations. Despite large differences between untreated M at matching experiments on Cannock Chase (Staffordshire), treatment effects were similar at both sites (cutting twice per year or cutting together with herbicide application were best). At Sourhope (Cheviot Hills) a different pattern of response was obtained, with cutting alone giving better results than treatments involving herbicide.
7The ratio of rhizome to frond dry mass per unit area was 10 : 1 in good conditions for growth, but the proportionate size of rhizomes could be reduced in poor growing conditions such as wet summers.
8Synthesis and applications. In order to develop a national control strategy the following must be considered: rhizome mass differs between sites and in response to control treatments; cutting twice per year is generally most effective; where cutting is impossible, herbicide treatment should be applied. Weather may affect rhizome mass, with wet years being detrimental. This raises the possibility that bracken may increase under the drier conditions that may occur with global warming.
Bracken Pteridium aquilinum L. Kuhn is a problematic weed world-wide, causing difficulties for managers of upland and marginal land in the UK (Marrs et al. 2000). Many attributes make control difficult but most workers identify the rhizome system as the major factor (Kirkwood & Archibald 1986; Marrs, Johnson & Le Duc 1998a; Martin 1976; Robinson 2000). Three strategies are commonly used for bracken control (Marrs, Johnson & Le Duc 1998b). One is mechanical control: fronds are damaged, usually by cutting or bruising (e.g. mechanical rolling). The aim is to damage the fronds in early summer, with maximum frond expansion and depletion of rhizome carbohydrate reserve, before translocation replenishes the reserves (Williams & Foley 1976). Cutting may be applied up to three times annually (Williams 1980).
Secondly, use of herbicide: the most successful herbicides kill frond-producing buds on the rhizome, with little immediate effect on carbohydrate reserves. Asulam [methyl (4-aminobenzenesulphonyl) carbamate] is most commonly used. On translocation from frond to rhizome it kills active and dormant buds (Veerasekaran, Kirkwood & Fletcher 1977, 1978). Asulam frequently achieves almost 100% frond reduction in the year following spraying, but in following years rapid recovery may occur and follow-up treatments are needed (Lowday & Marrs 1992). With lowered frond productivity, respiration and decomposition slowly reduce carbohydrate reserves.
Thirdly, inhibition by other vegetation: where bracken is dense, other vegetation is often degraded or may have disappeared. Bracken litter may aggravate this degradation. Following control treatment, restoration by litter disturbance and seeding or fertilization may be attempted. There is evidence that development of competitive vegetation can help in controlling bracken (Watt 1955; Lowday & Marrs 1992; Petrov & Marrs 2000).
While many methods are used to control bracken, individual experiments produce conflicting results. There is a need for an integrated national control strategy that takes account of the impact on rhizome performance. We have attempted to develop such a strategy.
developing an integrated strategy for bracken control
In developing a national control strategy we need to consider the effects of appropriate combinations of bracken control treatments in different situations, particularly with differing climates and differing vegetation restoration objectives. As bracken control and subsequent vegetation restoration is slow in the British uplands (Pakeman, Le Duc & Marrs 1997) these studies must be long-term.
We aimed to formulate such a strategy based on the results of seven experiments set up (1993–95) at four sites in different regions of the UK (Sourhope, Cheviot Hills; North Peak Environmentally Sensitive Area (ESA), Peak District National Park; Carneddau, Snowdonia National Park; Cannock Chase, Staffordshire). All areas had dense bracken but vegetation types differed (Table 1), and hence their management prescriptions. At two sites the desired outcome was grassland, and at the other sites Calluna heath. We used a standard experimental design throughout and applied the same bracken control treatments as main-plot treatments. For each site subsidiary treatments were applied at subplot level. Subsidiary treatments were chosen to test hypotheses concerning (i) vegetation end-points; (ii) vegetation restoration techniques; (iii) bracken control follow-up treatments. We examined the impact of these treatments on the rhizome system; effects on frond production have been reported elsewhere (Le Duc et al. 2000).
Table 1. Site vegetation description. The National Vegetation Classification (NVC; Rodwell 1991a,b, 1992) descriptions represent the pretreated condition; figures in parentheses were obtained with bracken left out of calculations. Fit (= goodness-of-fit) was given by TABLEFIT version 1·0 (Hill 1996), and is: > 80 very good, > 70 good, > 60 fair, < 60 poor
Acid grassland with moderate grazing. There are conifer plantations and wet areas near both sites
Acid grassland with high grazing (sheep and ponies). Up- slope from the experiment the vegetation changes, first to Ulex then Calluna heath
Antecedent Calluna/ Erica heath altered by bracken invasion. Grazing levels low. Patches of pine woodland nearby, heath more distant
Has extensive areas of Calluna heath with patches of grass heath, large pine plantations, ancient oak woodland and birch scrub, all bracken infested. No grazing
Bracken litter status
Moderate litter disappearing rapidly after treatment
Moderate litter disappearing rapidly after treatment
Extensive litter, often in discrete heaps that remain after treatments
Extensive, often continuous, litter beds remaining after treatment
Site differences within location
Sourhope 2 more base-rich than Sourhope 1, but prone to development of bare patches and invasion by other weeds. Suffers from trampling
Cannock 1 adjacent to oak–birch woodland; Cannock 2 has some pine invasion; Cannock 3 in danger of pine and birch invasion. Cannock 1 and 3 suffer from trampling
Site NVC class
Sourhope 1: U4e (U4e), fit 75 (77); Sourhope 2: U4a (U4a), fit 61 (64)
U4a (U4a), fit 74 (79)
U20 (H18), fit 77 (83)
Cannock 1: W16 (U2b), fit 78 (67); Cannock 2: W16 (U2b), fit 61 (55); Cannock 3: U2 (U2), fit 63 (62)
problems in studying rhizomes
The rhizome system comprises a massive underground network, with growth and senescence causing fragmentation and temporal change. Confusion over nomenclature exists in the literature. We define a bracken patch as an area of fronds comprising single or multiple genets and/or single or multiple plants. A genet has constant genetic make-up but may comprise many individuals (clones), i.e. independent physiological units (Daniels 1985). A plant is a single genet and a physiologically independent unit. These definitions concur with Birch (2002). Watt (1940) suggested the actual size of a bracken plant was usually quite small in comparison with its potential, rarely exceeding 20 m from rhizome apex to the senescent end. It has also been suggested that the extent of a single genet (perhaps many plants) may be considerable (Sheffield et al. 1993).
Lateral transfer of nutrients occurs across short distances (Tyson, Sheffield & Callaghan 1999) but can only occur within single plants. When sampling rhizomes it is possible only to obtain a broad view at patch scale; it is impossible to obtain plant-scale information without an archaeological approach.
The rhizome system presents problems for experimental design. Lateral translocation leads to non-independence of experimental units, increasing potential for type II errors. Increased plot size reduces this effect, and randomization techniques can be used to test the robustness of statistical results (Manly 1997).
The bracken rhizome is dichotomously dividing, comprising stem tissue of systematically varying internode length (length between branches; Watt 1940; Whitehead & Digby 1997). Watt defined three readily identifiable forms: long, intermediate and short shoots, depending on internode lengths of 15–40 cm, 2–15 cm and 0·5–2 cm, respectively. Although short shoots arise from long and intermediate ones, there is evidence that they can revert back again (Watt 1940). Short shoots are principally the frond-bearing components (Watt 1940), commonly bearing petiole scars and found near the soil surface. The main function of long and intermediate shoots is carbohydrate storage but they are also important in foraging, with an occasional minor role in frond bearing (Whitehead & Digby 1997). As the function of intermediate and long shoots appears more or less identical we do not differentiate between them. We considered two categories, the short shoots and the long shoots (including intermediate shoots), and refer to these using the functional terms ‘frond-bearing’ and ‘storage’ rhizomes (Williams & Foley 1976).
Williams & Foley (1976) monitored carbohydrates within fronds and rhizomes throughout the year, and suggested that rhizome reserves were maximum in early autumn. Thus, we made measurements in October.
Various measures have been used to assess the impact of control treatments on the rhizome, including length, diameter, internode length, number of buds, carbohydrate concentrations and biomass. However, two separate studies in Breckland, UK, have shown that the best predictor of frond performance, for a range of sites, was total rhizome biomass (Marrs, Johnson & Le Duc 1998a). In a small number of cases bud counts were significant, but length (m m−2) was not. We therefore confined our assessment to the effects on rhizome biomass
The impact on frond production differs between the two main methods of control, cutting and herbicide application. Cutting maintains frond density but length is reduced; herbicide reduces density significantly but length is maintained (Le Duc et al. 2000). The frond-bearing fraction of the rhizome system must reflect this difference. Therefore two aspects of response were examined: (i) total rhizome dry mass per unit area (M) and (ii) the ratio of frond-bearing to total dry mass (R). We also discuss the effects on rhizomes in relation to earlier work on fronds (Le Duc et al. 2000). Nomenclature follows Stace (1991).
Main bracken control treatments
Seven experiments were established in four different locations in the UK (Table 2). Two pairs of experiments (Sourhope 1 and 2; Cannock 1 and 2) were matched in experimental design, with identical treatment history except that Sourhope 2 was started 1 year after Sourhope 1.
Table 2. Description of split- or split-split-plot experiments to test bracken control and vegetation restoration methods. Entries for experimental designs represent: number of blocks/number of plots per block/number of subplots per plot/number of subsubplots per subplot. Codes are truncated from the right when lower experimental levels do not exist. Treatments described further in the text. Parentheses enclose treatment (or treatment start) date
Location Ordnance Survey map reference
North Peak ESA
NT 861 202
NT 846 210
SK 213 870
SH 690 711
SJ 976 200
SJ 987 181
SJ 987 178
Main treatments: (1) untreated control; (2) cut once per year; (3) cut twice per year; (4) cut in year 1 and sprayed with herbicide in year 2; (5) sprayed only; (6) sprayed in year 1 and cut in year 2.
Weed-wipe follow-up to sprayed only treatment (September 1996)
Weed-wipe follow-up to Sprayed only treatment (September 1996)
Subtreatment, vegetation restoration
Grass seed, 2 levels including control (August 1993)
Grass seed, 2 levels including control (August 1994)
Stock fence, 2 levels including control (January 1994)
Fertilize (April 1995) only or combined with grass seed (June 1995), 3 levels including control
Fertilize only or combined with Harrow, 3 levels including control (March 1994)
Fertilize only or combined with harrow, 3 levels including control (March 1994)
Litter burn, 2 levels including control (March 1995)
& follow-up vegetation restoration
Re-fertilize only (May 1996)
Subsubtreatment, vegetation restoration
Cluna seed (either with brash or litter), 3 levels including control (November 1993)
Calluna seed (brash), 2 levels including control (December 1996)
& follow-up bracken control
Spot-spray, 2 levels including control (August 1997)
Rhizome harvesting year
First harvest 1998, second 1999–2000
The bracken control treatments were applied to main plots within spatially replicated blocks, as their implementation required all-terrain vehicle (Kawasaki quad) equipment. Vegetation restoration treatments were site-specific, and were applied within main plots at split-plot or split-split-plot levels. All the main-plot bracken control treatments, and split- and split-split-plot vegetation restoration treatments, were allocated randomly, in balanced designs, and included untreated controls. Bracken control follow-up treatments were included in some experiments to assess standard bracken control procedures (Table 2). Bracken control treatments were: (i) untreated control; (ii) cut once per year; (iii) cut twice per year; (iv) cut year 1, sprayed with herbicide (asulam) year 2; (v) sprayed once only; (vi) sprayed year 1, cut year 2. Cannock 3 was exceptional in having only two main treatments (untreated control and cut twice per year). Combination treatments had single applications of the component treatments. Generally, single cuts took place in June, second cuts and herbicide application in August.
Vegetation restoration and follow-up
As there were site vegetation differences (Table 1), treatments varied. Sourhope 1 and 2: grass seeding (± grass seed) to improve sward quality and prevent bare patches (subtreatment experimental level). Sourhope 1 and 2: weed-wiping, standard for follow-up control, tested by comparison between two experiments (Sourhope 1 wiped, Sourhope 2 not wiped). Carneddau: spot-spraying (± spot spray) for follow-up bracken control (subsubtreatment level). Carneddau: fertilizer application, alone and combined with grass seeding (three treatments levels, including control, at subtreatment level). North Peak: two methods of Calluna seeding (brash or litter, three treatment levels including control) for restoration of heath (subsubtreatment level). North Peak: ESA prescription rate grazing (0·5 sheep ha−1) vs. no grazing (two treatments at subtreatment level). Cannock Chase 1 and 2: weed-wiping for follow-up bracken control (comparison between Cannock 1 with and 2 without). Cannock Chase 1 and 2: fertilizer application with and without litter disturbance, by harrowing (three treatment levels, including control, at subtreatment level). Cannock Chase 3: Calluna seeding by brash (two treatment levels including control, at subsubtreatment level). Cannock Chase 3: litter disturbance by burning (two treatment levels including control, at subtreatment level).
The grass seed mix included Festuca ovina, Agrostis capillaris, Poa pratensis and Rumex acetosa. The Calluna seed treatments included Agrostis castellana as nurse crop at North Peak and A. capillaris at Cannock 3. The experiments (except Cannock 3) have also been described elsewhere (Le Duc et al. 2000).
Rhizome sampling took place in October over 3 years: Cannock was sampled in 1998, Sourhope in 1999 and North Peak in 2000 (scheduled for 1999, abandoned due to waterlogged soil). Carneddau was sampled in 1998 (the Carneddau 1998 data set), and was followed in 1999 by a sample of 50% of plots, chosen at random, and the complementary 50% in 2000; this was to attempt a sampling quality check through time, and a prospective test for temporal changes. The two 50% samples were combined to form a second complete data set (Carneddau 1999 & 2000).
Two samples were taken per subplot, for split-plot experiments, and one per subsubplot in others (Table 2). Samples comprised the rhizome content of a 0·5 × 0·5-m pit with varying depth according to rhizome presence. Pits were located centrally to a 1 × 1-m quadrat randomly selected from a 1-m grid within each sub(sub)plot. Quality control of pit size was achieved by requiring that a quadrat 0·495 × 0·495 m should pass to the bottom on completion. A non-systematic sample of pit depths gave median values of: Sourhope 1, 24 cm (range 16–38 cm); Sourhope 2, 40 cm (25–47 cm); Carneddau, 35 cm (25–50 cm); Cannock 1, 48 cm (15–80 cm); Cannock 2, 63 cm (25–100 cm). A total of 580 pits was dug.
Samples were washed with water (6 bar) to remove soil and remaining fragments of dead rhizome, and dried at 80 °C. Dried samples were weighed giving total rhizome dry mass (M, kg m−2), then separated into storage and frond-bearing fractions. Both fractions were reweighed, giving masses ms and mf, respectively. A small portion of fragments, of indeterminate type, was discarded. We assumed M=ms + mf, and fraction weights were adjusted accordingly.
Finally, the ratio R=mf/M was calculated. Thus changes in M and/or R indicate a range of functional changes in emphasis in the rhizome system: low R indicates a greater portion of storage rhizomes, and high R vice versa.
Experimental designs (Mead 1988) are described in Table 2. Each experiment was analysed independently (proc anova; SAS 1989), using the response variables total dry mass, M, and ratio, R. The latter required transformation prior to testing. The usual logarithmic transformation did not improve the distribution (Anderson–Darling test; Stephens 1974). However, the arcsine transformation (Sokal & Rohlf 1995), used, for example, in morphometric work with molluscs (Tokeshi, Ota & Kawai 2000), was satisfactory. To test for experimental differences at regional level, we analysed replicated experiments (Sourhope 1 and 2, and Cannock 1 and 2) using the method of Mead (1988).
A significant weed-wiping effect would give significant interaction between experiment and treatment when testing paired experiments. However, this would be confounded by experiment. Hence a significant interaction would necessitate a meta-analysis, with the two pairs of experiments combined, to correctly identify the cause.
To check for increased type II errors, due to non-independent sample units, we carried out a Monte Carlo randomization procedure (Manly 1997). Ninety-nine permuted data sets (restricted by design) were produced using shufflerows (PopTools DLL running in Microsoft EXCEL; Hood 2000). These data sets were analysed using anova, thereby producing 99 estimates of F-ratio to compare with the original data. Hence an absolute test on F-ratios, with a type I error of α = 0·01, was obtained.
rhizome performance in untreated bracken
There were considerable differences in rhizome characteristics in untreated plots, both between sites and years (Table 3). Cannock 1 had the greatest M, 5·1 kg m−2. The three Cannock experiments (measured in 1998) showed considerable variability, Cannock 1 having twice the total rhizome mass of the others, although R was similar in all three. In contrast, both Sourhope sites had similar M (2·9–3 kg m−2). Carneddau was the only experiment sampled in different years and revealed a lot of variation; M was greatest in 1998 with considerable reduction in subsequent years. However, little evidence exists for annual change in R at Carneddau. Peak had an unexpectedly small value of M, but not as small as for Carneddau in 2000.
Table 3. Mean dry mass of rhizomes in plots untreated at every experimental level, by experiment and sample year. M is total rhizome dry mass; ms is dry mass of the storage rhizome fraction; mf is dry mass of the frond-bearing rhizome fraction. Standard error of means in parentheses (values for Carneddau 1999 & 2000 not available separately)
Dry mass fraction (kg m−2)
R (= mf/M)
The R ratio varied regionally: Cannock had greatest R (≥ 0·4), with Carneddau next (0·3–0·4), then Peak (∼0·3) and Sourhope (0·24–0·3) the smallest (Table 3). The range of M was much greater than that reported by others (Table 4), the value for Cannock 1 being much larger than most of the others. Results from these upland sites were much larger than those reported for a lowland site (Marrs, Johnson & Le Duc 1998a). The latter were measured in April but differences should be minimal (Williams & Foley 1976). Daniels's (1981) figure, > 3·0 kg m−2, is nearest to those we measured.
Table 4. Estimated values for total rhizome dry mass, M, and dry-mass ratio of frond-bearing to total rhizomes, R, reported by other workers for untreated bracken
M (kg m−2)
Estimated from shoot dimensions, assuming constant densities.
Interpolated from graph of dry masses.
Fresh weight ratio values reported, thus uniform mass loss on drying assumed.
Results from the randomization procedure show that for all experiments, apart from Cannock 3, the pattern of significance obtained for each treatment level was similar to that of the original anova (Table 5). Thus, apart from Cannock 3, the assumption of plot independence was justified and the results obtained using standard anova were satisfactory. At Cannock 3 anova detected significance for subsubplot treatment but no effect was found with randomization. Cannock 3 subsubplots are smaller than elsewhere (30 m−2 compared with ≥ 50 m−2). It appears that the lower plot size had led to increased non-independence between subsubplots, leading to a higher type II error rate and reduced precision.
Table 5. Summary of anova results by experiment and response variable. Cannock 2 produced no significant results on its own (data not shown). Results shown if P < = 0·05 or, in a few cases, where randomization test P < = 0·05
Source of variation
Randomization test P
All experiments have the same set of main treatments at six levels (see text) except Cannock 3 with only two levels.
Generally, bracken control treatments affected M significantly but there were exceptions (Table 5). Sourhope 2 and Cannock 2 showed no significant effect. At Cannock 2 this might be explained by the large variation found there (coefficient of variation, V= 44% compared with Cannock 1 V= 25%). A smaller difference was found between Sourhope 2 and Sourhope 1 (V = 32% and 26%, respectively). Cannock 3 (single main treatment; Table 2) produced a treatment effect only when combined with subsubtreatment (Calluna brash).
In three experiments, Sourhope 2, Peak and Carneddau 1999 & 2000 data, a significant change in R was seen in response to the main treatments. Also in the Carneddau 1999 & 2000 data set, several significant effects on M were detected, with respect to bracken control treatments, subtreatments and interactions, and for bracken control treatments on R. This latter effect suggests that, in response to treatments, changes in allocation pattern had occurred.
Analyses of main treatment effects for matching pairs of experiments (Sourhope 1 and 2; Cannock 1 and 2) produced significant results for M only.
bracken control treatment effects by experiment
Effects on M
The effects of bracken control treatments on total rhizome mass (M) were regionally specific (Fig. 1). At Cannock (largest M in untreated plots) some treatments had little or no effect, i.e. cut once per year, spray, and spray and cut at Cannock 1. At all other sites bracken control treatments reduced M. The two Sourhope experiments responded in the same way, with all treatments reducing M, although none was significant at Sourhope 2.
In most experiments cutting twice per year produced the greatest reduction in M, but not with Sourhope 1 and Cannock 2. At those experiments cut and spray was as effective, with the advantage that treatment for only 2 years gave the same effect as cutting twice per year for 4 or 5 years (Table 2). Cutting once per year was less effective overall, having no effect at Cannock, but was as good as any treatment at Sourhope.
The single experiment analyses produced three significant anova results for R (Table 5), namely for Sourhope 2, Peak and Carneddau 1999 & 2000 data. For Peak and Carneddau 1999 & 2000 some treatments reduced R relative to untreated plots, indicating a significantly lower mass of frond-bearing rhizomes in proportion to overall mass (Fig. 2), a condition reported to be present close to invading fronts.
At the Peak experiment, cut and spray (R = 0·135) and spray and cut (0·127) gave values significantly lower than untreated ones. However, the treatments most successful in reducing mass M, cut once and twice per year (R = 0·266 and 0·283), did not affect R (Fig. 2). At Carneddau 1999 & 2000, spraying reduced R significantly compared with the untreated control (R = 0·161 vs. 0·299). At Sourhope 2 the highest R was for cut and spray (0·263) and the lowest for cut once per year (0·216).
combined matching experiments
No significant experiment effect was detected for either pair of matching experiments, Sourhope 1 and 2 or Cannock 1 and 2. However, there was a significant interaction between experiment, bracken control treatments and subtreatment (grass seeding) at Sourhope for M (Table 6; see vegetation restoration below).
Table 6. Rhizome total dry mass (M, kg m−2) responses to experiment × treatment × subtreatment interactions (in bold) for two experiments at Sourhope. For comparisons, 2 × standard errors of differences are: experiment = 0·71; treatment = 0·40; subtreatment = 0·19; experiment × treatment = 0·57; experiment × subtreatment = 0·27; treatment × subtreatment = 0·47; experiment × treatment × subtreatment = 0·67
Untreated × no seed
Untreated × grass seed
Cut once per year × no seed
Cut once per year × grass seed
Cut once per year mean
Cut twice per year × no seed
Cut twice per year × grass seed
Cut twice per year mean
Cut year 1, spray year 2 × no seed
Cut year 1, spray year 2 × grass seed
Cut year 1, spray year 2 mean
Spray × no seed
Spray × grass seed
Spray year 1, cut year 2 × no seed
Spray year 1, cut year 2 × grass seed
Spray year 1, cut year 2 mean
Bracken control treatment effects
Analyses of the combined experiments (Sourhope 1 and 2, Cannock 1 and 2) emphasized the effects of treatment on M from the single experiment analysis (Table 5). The overall pattern of response at Sourhope was quite different from Cannock. At Sourhope the best treatment was cut once per year, giving a mean M of 1·14 kg m−2 compared with 2·87 kg m−2 (significantly the highest) for untreated plots. At Cannock 1 and 2 (combined) cut and spray had the lowest mean M, 2·41 kg m−2, cut once per year was the highest (4·99 kg m−2), with the untreated control having a mean not significantly smaller (4·60 kg m−2).
The combined analyses did not detect any bracken control treatment effect on R. This was despite the highly significant effect at Sourhope 2 when tested alone.
There were no significant interactions between experiment and main treatments. Thus, on a local scale (e.g. Sourhope or Cannock), bracken control treatments produced similar effects when applied to different stands.
vegetation restoration and follow-up treatments
There were no changes in R associated with any vegetation restoration or follow-up treatments. Of the 10 statistical tests for treatments (see the Methods) those involving weed-wiping did not produce significant differences in M.
Grass seeding: Sourhope
There was no significant effect at single-site level. When the experiments were analysed together, a significant effect was found for M (randomization test supported a marginally significant result) for the interaction between experiment, treatment and subtreatment (Table 5). The rhizome system responded differently at each site for two treatments, cut once per year and spray (Table 6); at Sourhope 1 seeding was associated with smaller M than for unseeded plots (for both main treatments) but for Sourhope 2 the reverse was found.
There was no effect on M in the Carneddau 1998 data; but in the 1999 & 2000 data there was a significant effect (M = 1·09 kg m−2 vs. 0·87 kg m−2 with spot-spraying). Spot-spray had no significant interaction with any bracken control treatments, appearing to be equally effective with any initial treatment.
Fertilizer and grass seeding: Carneddau
There was no detectable effect in the Carneddau 1998 data. However, in the Carneddau 1999 & 2000 data an interaction with the spot-spraying subsubtreatment was found (Table 5). When combined with spot-spraying, fertilizer plus seed application produced a significantly smaller M (Fig. 3b).
Callunaseeding and grazing: North Peak
The only effect on M was for seeding combined with grazing subtreatment (Table 5). There was a difference between no seed (1·23 kg m−2) and Calluna litter addition (1·59 kg m−2) when fenced. Litter application combined with grazing also gave a significantly smaller M (Fig. 4a).
As a single treatment, grazing had no detectable effect on rhizome dry mass. When combined with Calluna seeding, a complex interaction was obtained for M (Fig. 4a).
Fertilizer and harrowing: Cannock 1 & 2
This produced a significant effect on M at Cannock 1, and for the combined Cannock experiments, but not for Cannock 2 (Table 5). The interaction of fertilizer with disturbance led to a significantly smaller M than fertilizer on its own or no subtreatment (Fig. 3a).
Callunaseeding: Cannock 3
The randomization test result implied that the apparent significant result for this subsubtreatment was unsafe (Table 5). Apparently significant interactions were found (randomization test P= 0·07) both with the bracken control treatment (cut twice per year) and with litter burning subtreatment. Litter burn without Calluna gave a greater M than for burn combined with brash or no subtreatment (Fig. 4b). The apparent interaction with the main treatment was that combined cutting and Calluna brash application led to a smaller M than for other interactions. The no treatment interaction was smaller than other combinations.
Litter burning: Cannock 3
The subtreatment produced a possible significant effect (randomization test P= 0·07) on M when interacting with subsubtreatment Calluna brash (Table 5). The effect was a greater M for litter burn without Calluna than for burn combined with brash or no subtreatment (Fig. 4c).
variation of the rhizome system
Other workers have reported variations in M within a single stand and from year to year (Table 4); a much greater variation is reported in this work. This reflects our objective of studying bracken performance across a range of environments reflecting bracken's broad ecological amplitude (Grime, Hodgson & Hunt 1988). Values of R reported by others are rarely as high as the maximum reported here. Most low values (Table 4) were obtained near actively invading bracken fronts. The range of R reported here implies that the experiments occupied areas of bracken stands distant from invading fronts, reflecting the deliberate intention to avoid such edge effects.
At Carneddau variation in productivity from year to year was considerable, with successive reductions in rhizome biomass from 1998 to 2000. While we have no experimental evidence we hypothesize that this might have resulted from recently very wet years, a view supported by Poel's (1961) work. Meteorological data for Valley, Anglesey (40 km from Carneddau), for the 1999 and 2000 growing seasons (May–September inclusive) gives total rainfall greater than the long-term average by 74 mm (24%) and 52 mm (17%), respectively. This wet period followed a dry growing season (1998) with 39 mm (13%) below average rainfall. Estimated annual variation in rhizome biomass at Carneddau (Table 3) might be explained by the dry summer followed by two very wet summers.
Reduced productivity of bracken under high soil moisture contradicts the assumption (Pakeman et al. 1994) that increasing soil moisture leads to increasing dry matter production. However, bracken is sensitive to reduced aeration (Poel 1961), thus the relationship between bracken productivity and soil moisture is more probably curvilinear, with an optimum response associated with moderate soil moisture, and influenced by soil texture. Apparently at Carneddau this optimum relates to drier conditions than in 1999 and 2000.
If this response holds for other sites, measurements for Cannock (sampled in 1998) might relate to highly productive conditions, Sourhope (sampled in 1999, after one wet growing season) intermediate conditions, and Peak (sampled in 2000, after two wet seasons) poor conditions. The waterlogging found at Peak in 1999 supports this view, and may be responsible for the very low M in untreated plots (Table 3). Cannock may be less tightly coupled to summer rainfall, as the soils are sandy and less prone to waterlogging. Thus the regional variation may, in part, have resulted from differing sampling years. To test this hypothesis annual sampling (suggested by Poel 1961) across sites is now in progress. Climate is a major factor controlling bracken production; high soil moisture deficits reduce productivity (Hollinger 1987). It has been suggested that, for certain climate change scenarios, infestations will increase and be more difficult to control (Pakeman & Marrs 1996). However, climate warming with increased summer rainfall, predicted for some regions (Naden & Watts 2001), may suppress growth according to the argument presented above.
Intersite differences can be attributed partly to within-site variations. Large variations at Sourhope 2 and Cannock 2, resulting in non-significant treatment effects, may have different causes. The variation at Sourhope 2 might result from a degree of senescence, thus explaining the low R in untreated plots. At Cannock 2 there is a systematic variation in soil depth across the experiment, with shallower soil supporting shorter fronds and giving shallower rhizome pits. Large within-site variation in rhizome mass may result in the large variation in bracken control efficacy often reported.
We accept there is considerable speculation in this discussion of environmental controls of bracken rhizomes and further experimental work is needed to test these hypotheses.
bracken control treatments
The main requirement for good bracken control is systematic reduction of carbohydrate reserves, hence frond production potential. Against this criterion the effectiveness of treatments under investigation varied considerably.
Generally, the most successful treatment was cutting twice a year, but there were exceptions; the best at Sourhope 1 was cut once per year, and at Cannock 2 cut and spray. It appears that a small amount of effort is needed for control if the bracken is in a low productivity state (e.g. Sourhope 1). Here, more vigorous treatment might be counterproductive, inducing a damage-response mechanism. Highly productive bracken appears to be resilient to damage at any level of treatment and requires a great amount of effort to reduce its productivity (e.g. Cannock 1). This hypothesis remains to be tested.
Treatments under investigation were either continuous (cut once or twice per year) or one-off (e.g. spraying). It is considered good practice (Robinson 2000) to carry out follow-up treatment after spraying; this was tested in two ways. The first method, weed-wiping carried out in 1996 at one of each of the paired experiments (Sourhope 1 and Cannock 2), did not have a significant effect according to analysis of the paired experiments. The second follow-up treatment, spot-spraying (Carneddau 1997), was not effective in the first sampling, 1998, but was in 1999 and 2000. It is not possible to say whether the spray with follow-up combination was different from cutting treatments alone because the interaction between main treatments and subsubtreatment was not statistically significant.
vegetation restoration treatments
Restoration treatment, proposed as an additional strategy for bracken control (Lowday & Marrs 1992; Marrs, Johnson & Le Duc 1998b; Petrov & Marrs 2000), changed M in some situations. Generally, where vegetation had been improved in some way M was reduced, e.g. by fertilizing and seeding (Carneddau), Calluna seeding and grazing manipulation (Peak) and fertilizing and harrowing (Cannock). Presumably this results from increased competition between ground vegetation and bracken, essentially following the inhibition model of succession (Connell & Slatyer 1977).
Functionally, we considered rhizomes to comprise two main anatomical structures. First is the long- and intermediate-shoot combination, the main functions being carbohydrate storage and foraging; second is the frond-bearing shoot. As frond-bearing rhizomes can develop into new long/intermediate shoots, it is likely that bracken can respond to prevailing conditions by shifting allocation between the two systems. We detected such changes by examining the ratio of frond-bearing to total rhizome dry mass, R. It would be surprising if plastic responses of the rhizome system did not occur, considering the large difference in values of R detected here and by others (Table 4).
Others (Watt 1940; Whitehead 1993) have shown that, at advancing fronts where foraging is a major function, frond-bearing rhizomes form only about 15% of the total rhizome dry mass. In mature stands, some distance from the front (> 20 m according to Watt 1940), the main function is carbohydrate production and frond-bearing rhizomes increase in proportion to about 30%. We report values up to 42% from highly productive sites.
Of the seven experiments reported only three returned significant bracken control treatment effects on R: Sourhope 2, Peak and Carneddau 1999 & 2000. The data suggest that these sites represent low productivity stands or phases. Low carbohydrate reserves might trigger functional plasticity to optimize light harvesting under the poor conditions experienced in recent summers. The Peak and second Carneddau data sets, 1999 & 2000, were collected later in this period, and certain treatments (spray at Carneddau and combined cutting and spraying treatments at Peak) reduced R significantly, giving values similar to those found at invading fronts several years after treatment.
comparison with frond performance
For untreated bracken it is possible to compare frond and rhizome dry masses using compatible data (Table 3; Le Duc et al. 2000table 2); the Carneddau 2000 data were not used as n= 1. The correlation obtained is not significant, which appears to be due to low rhizome dry mass at Peak after two seasons of poor weather (Fig. 5). The shift between the 1998 and 1999 results for Carneddau supports this theory. Omitting the Peak data gives a better result (r = 0·93, d.f. = 4, P= 0·008). Thus, under good growing conditions, the regionally consistent ratio between rhizome and frond dry mass is approximately 10 : 1. This figure contrasts with unpublished data from R.J. Pakeman & R.H. Marrs, who found an average of about 2 : 1 during 1990–92 for sites on the North York Moors (Pakeman & Marrs 1993) and in Breckland (Pakeman & Marrs 1994), UK.
From evidence presented here, there appear to be circumstances of exceptionally poor growth conditions when the bracken rhizome can become seriously depleted. Under those conditions functional changes occur in the rhizome system and R becomes responsive to the effects of control treatments, as was seen at Peak and Carneddau in 1999.
From frond data the most effective treatments were those combining cutting and spraying (Le Duc et al. 2000), as expected because asulam kills the buds for next year's fronds. However, with rhizomes the only site where cut and spray appeared to give a better result (not statistically significant) than cut twice per year was Cannock 2, and at Cannock 1 these treatments were not distinguishable statistically. Thus cut and spray might be preferable over other treatments at high productivity sites.
conclusions for strategic bracken management
An important conclusion from this study is that it is essential to take into account the rhizome system within a control strategy. Large variations in rhizome mass between- and within-sites, and between years at the same site, were found. The rhizome mass may be as much as 10 times the dry mass of the fronds. Taking the results for rhizomes and fronds (Le Duc et al. 2000) together, three treatment combinations are recommended for bracken control. Annual cutting should be used for accessible sites and should be twice a year, especially if the rhizome biomass is high. On less accessible sites with very productive bracken, a single cut followed by spraying with asulam in the following year is recommended, but retreatment, using spot-spraying, will be necessary after several years. Where the rhizome biomass is low, herbicide spraying alone is often very effective but continued follow-up treatment is essential and should be carried out annually if possible (Robinson 2000).
However, even if a stand is senescent there is a threat of rejuvenation in response to control treatment. Furthermore, subtle differences in response may occur at apparently similar sites; seed application may assist in control at one site but assist recovery at another. Moreover, it may be possible to take advantage of weather conditions, especially under continued waterlogging; here the rhizome system may be weakened sufficiently for very effective control by spraying.
The work was supported by the Department for Environment, Food & Rural Affairs. We are indebted to landowners and management for permission to use experimental sites, particularly Harry Sangster, Neil Taylor, Jeremy Archdale, John Morgan, Elfyn Jones, Roger Hill and Sue Sheppard. Many individuals gave invaluable field assistance, including students from Chester College and the University of Liverpool. Thanks are also due to Dr Hugh McAllister for useful discussions and anonymous referees for useful comments.