Control of Molinia caerulea on upland moors


  • R. H. Marrs,

    Corresponding author
    1. Applied Vegetation Dynamics Laboratory, School of Biological Sciences, University of Liverpool, PO Box 147, Liverpool L69 3GS, UK; and
    Search for more papers by this author
  • J. D. P. Phillips,

    1. Applied Vegetation Dynamics Laboratory, School of Biological Sciences, University of Liverpool, PO Box 147, Liverpool L69 3GS, UK; and
    2. The Heather Trust Ltd, The Cross, Kippen, Stirlingshire FK8 3DS, UK
    Search for more papers by this author
  • P. A. Todd,

    1. Applied Vegetation Dynamics Laboratory, School of Biological Sciences, University of Liverpool, PO Box 147, Liverpool L69 3GS, UK; and
    2. The Heather Trust Ltd, The Cross, Kippen, Stirlingshire FK8 3DS, UK
    Search for more papers by this author
  • J. Ghorbani,

    1. Applied Vegetation Dynamics Laboratory, School of Biological Sciences, University of Liverpool, PO Box 147, Liverpool L69 3GS, UK; and
    Search for more papers by this author
  • M. G. Le Duc

    1. Applied Vegetation Dynamics Laboratory, School of Biological Sciences, University of Liverpool, PO Box 147, Liverpool L69 3GS, UK; and
    Search for more papers by this author

Rob Marrs, Applied Vegetation Dynamics Laboratory, School of Biological Sciences, University of Liverpool, PO Box 147, Liverpool L69 3GS, UK (fax +44 151794 4940; e-mail


  • 1Molinia encroachment has been viewed as a major threat to moorland conservation in the UK and elsewhere in Europe. In England and Wales agri-environment schemes are in place that aim to reduce Molinia caerulea and encourage the development of dwarf shrub vegetation. We tested a range of management treatments to achieve these objectives in two regions (the North Peaks and Yorkshire Dales) in England.
  • 2Within each region, the same experiment was carried out on two types of moorland vegetation, Molinia-dominated ‘white’ moorland and a mixture of Molinia and Calluna vulgaris‘grey’ moorland. Burning, grazing and herbicide (glyphosate) treatments were applied in factorial combination at each of the four sites (two regions × two moor types). The responses of both vegetation and individual species were assessed. In addition, on the white moors two techniques for Calluna re-establishment were investigated, (i) removal of Molinia litter by raking and (ii) application of Calluna seed.
  • 3The data were analysed using a combination of univariate and multivariate analysis of variance to identify trends in this complex data set.
  • 4The only treatment that had consistent effects in the univariate analysis of variance was glyphosate application, which had similar effects on Molinia at all study sites. There was little difference between the use of low and high application rates (0·27 and 0·54 kg ai ha−1). There was little impact of herbicide use on other moorland species. Some species were affected on some sites in some years, but there were no consistent effects. Tentative identification of species that responded positively, negatively and erratically to glyphosate application was made.
  • 5Greater Calluna seedling densities were found in the plots where herbicide was applied, the Molinia litter was removed and seed was added. However, after initial colonization, there was a reduction in Calluna seedling densities as the Molinia recovered. This indicated that disturbance, seed addition and follow-up management are required for successful Calluna establishment.
  • 6There were significant differences in community response between both the regions and moorland types. The Dales had a relatively greater contribution of grassland species than the Peaks, where the grey site had a relatively greater dwarf shrub component.
  • 7Burning had little effect on community composition but both grazing and herbicide application had important effects. Grazing of the grey sites, even at the very low levels used in this study, tended to push the communities towards bog–moorland vegetation, but little effect was found at the white sites. Glyphosate treatment tended to push communities towards acidic grassland at the Dales grey site but not at the Peaks. Successional change was also noted, with marked change between the third and fourth year and again between the fifth and six year. Grey sites showed the greatest temporal change.
  • 8Synthesis and applications. In terms of Molinia control and subsequent restoration of dwarf shrubs, there was marked variability of response between ‘apparently similar’ vegetation types in different regions. There were abrupt temporal changes taking place some years after treatment application and a significant length of time was required for change to be detected. Managers need to obtain a greater knowledge of initial floristic composition before starting the restoration process, be prepared to accept multiple outcomes of response (acid grassland vs. dwarf shrubs), be prepared for a long-term monitoring process and perhaps the inclusion of additional treatments for continued Molinia control (application of selective graminicides) and dwarf shrub restoration (disturbance and seed addition treatments).


Molinia (purple moor grass) abundance has increased at the expense of Calluna (heather or ling) since the start of the industrial revolution in upland areas of England and Wales and in parts of southern and western Scotland (Chambers, Dresser & Smith 1979), and anecdotal accounts indicate that increases have continued over the last 20–30 years. Increases in Molinia have also been found elsewhere in Europe, where large areas of Dutch heathlands have been colonized by Molinia at the expense of dwarf shrub vegetation (Heil & Diemont 1983; Diemont & Heil 1984).

These observed shifts in vegetation composition from Calluna to Molinia have been variously attributed to inappropriate burning and/or grazing regimes (Grant, Hunter & Cross 1963; Miller, Miles & Heal 1984; Grant & Maxwell 1988) and increased atmospheric nitrogen and sulphur deposition, especially in Holland (Chambers, Dresser & Smith 1979; Heil & Diemont 1983; Diemont & Heil 1984; Lee et al. 1987; Hogg, Squires & Fitter 1995; Roem, Klees & Berendse 2002). In the UK, many moorland areas are now eligible for grant payments to restore Calluna cover under agri-environment schemes, notably the environmentally sensitive areas (ESA) and countryside stewardship (CS) schemes (MAFF 1993, 1996). In such schemes the policy objectives are to increase the dwarf shrub component of the moorland and reduce the Molinia cover (Bardgett, Marsden & Howard 1995). Elsewhere in Europe where this problem occurs, the control of Molinia and restoration of heathland is a nature conservation issue (Heil & Diemont 1983; Diemont & Heil 1984).

In order to achieve the objectives within the framework of an agri-environment agreement, it is essential to take account of all users of moorland areas, including farmers, managers of sporting estates, conservationists and those who appreciate the aesthetic quality of the uplands. While there is a desire to maintain, or create, appropriate Calluna-dominated moorlands for wildlife, sporting and amenity interests (Tharme et al. 2001), there is a need to provide grazing livestock with the maximum possible level of nutrition from this indigenous vegetation throughout the year (Maxwell et al. 1986). To achieve this, there must be an adequate mixture of Molinia and Calluna (Maxwell et al. 1986). Molinia is a highly digestible species but, as it has a relatively short growing season (late April–late August), it only provides summer feed. Calluna, although having a lower nutritional quality, provides winter food (Grant & Maxwell 1988). Summer grazing can keep Molinia in check (Hulme et al. 2002). Any management prescription should not attempt the total eradication of Molinia, but achieve a reduction in its dominance and an increase in the dwarf shrub component of the vegetation, especially Calluna. Moreover, as a vegetation mosaic, there are complex interactions between vegetation communities and livestock utilization (Palmer & Hester 2000).

Once Molinia has been successfully controlled, and the litter layer is not inhibiting germination, Calluna may need to be established artificially. Although establishment can occur naturally from the seed bank if there are enough seeds present, experience suggests that litter must be disturbed and propagules added (Milligan 1998; Todd et al. 2000).

The limited information available on Molinia control is mainly from studies either in single locations where only one control method has been investigated, or conducted over a relatively short period (King & Davies 1963; Phillips 1995; Grant et al. 1996; Todd et al. 2000). Most studies have concentrated on Molinia control rather than on the effect on other members of the plant community. The present multi-site study was designed to provide generic information (Pywell et al. 2002) to evaluate control strategies for Molinia at two contrasting sites in two different upland regions in England (the Peak District and Yorkshire Dales). In these two regions there was a perceived policy objective to reduce Molinia and increase Calluna; moorland was eligible for either ESA or CS support schemes. The contrasting sites reflected different types of moorland, a ‘white’ moorland, where the vegetation is predominantly Molinia, and a ‘grey’ moorland, where the vegetation is a mosaic with some Calluna remaining in the vegetation (Thomas 1951; Phillips 1989). At each site a factorial combination of burning, grazing and herbicide applications was evaluated. Additional treatments to increase Calluna establishment in dense Molinia stands were also considered; these treatments were factorial combinations of (i) removal of Molinia litter and (ii) addition of material containing Calluna seed. The effects of the treatments on species composition and structure of the vegetation was assessed using both univariate analysis of variance and multivariate analysis of variance using constrained ordinations, and tested by Monte Carlo permutation tests (Ter Braak & Šmilauer 1998). While the aim was to inform policy makers in the UK on the most appropriate management prescriptions, the studies should by generally applicable for conservation management of Molinia throughout Europe.

Following common practice Molinia caerulea and Calluna vulgaris are referred to by their generic names, otherwise nomenclature follows Stace (1997) for higher plants and Corley & Hill (1981) for bryophytes.


study areas

The same experiment was carried out in (i) the North Peak District (Peaks) and (ii) the Yorkshire Dales (Dales), on moorland being managed by farmers within the Department for Environment and Rural Affairs (Defra) agri-environment schemes. The North Peak District is designated as an ESA and the Yorkshire Dales are eligible under the CS scheme (MAFF 1993, 1996). In both regions, Molinia encroachment on to moorland had been identified as a problem. A white and grey moorland was selected in each region, giving four experimental sites in total: two regions (Peaks/Dales) × two white/grey moorland types (site details; Table 1). The experimental sites used in this study were typical of Molinia-dominated sites in Britain, with a range of standing crops (4–5·3 t ha−1 ; Todd et al. 2000) comparable with those of Pearsall & Graham (1956).

Table 1.  Site locations and grazing prescriptions of the experimental sites in the two regions of England. ESA, designated as an environmentally sensitive area; CS, eligible for the countryside stewardship scheme
RegionAgri-environment schemeGrid referenceLongitude and latitudeMoorland typeDetails of grazing prescriptions
North PeakESANR 17 031°44′W, 53°31′NWhite0·87 ewes + followers ha−1 1 January−31 December
NR 17 031°44′W, 53°31′NGrey
DalesCSSD 82 912°16′W, 54°19′NWhite1·5 ewes + lambs ha−1 1 April−31 July
SD 82 912°16′W, 54°19′NGrey 

experimental design

At each site, two areas were demarcated to form experimental blocks of c. 4000 m2 each. In March 1995, within each block one half (c. 2000 m2) was selected randomly and the vegetation burned (burning = main plot treatment), the remainder was left unburned. Within each of the burning treatments, three sheep grazing subtreatments were set up (existing grazing regime, summer-only, ungrazed); this was achieved through the designation of three paddocks (each c. 600 m2) in both the burned and unburned areas. Grazing treatment was allocated to paddocks randomly. One paddock was a designated open area that allowed free access by sheep and cattle under the agreed management prescriptions for the ESA/CS scheme agreement for the area (Table 1); the second paddock was fenced but was opened at both ends to allow free access by livestock at ESA/CS scheme levels in summer (15 April−15 October) and closed in winter, giving a summer-only grazing treatment; the third paddock was fenced to exclude livestock. However, although the grazing treatments were applied in accordance with this plan, at all sites the agri-environment scheme grazing pressure was concentrated in the summer and there was little winter grazing. Thus the two grazing treatments were similar, although both could be contrasted with the ungrazed treatment.

Within each of the grazing subtreatments there were three herbicide application subsubtreatments (two rates of glyphosate plus an untreated control) applied to plots of 10 × 10 m separated by 2-m pathways. The herbicide treatment (glyphosate; Roundup Biactive, Monsanto, High Wycombe, UK) was applied in July 1995 with a knapsack sprayer at two rates, (i) high rate, 0·54 kg ai ha−1 (1·5 L product ha−1) and (ii) low rate, 0·27 kg ai ha−1 (0·75 L product ha−1), and compared with an unsprayed control.

On both white moorland experiments, a subsidiary experiment was also done. Two further treatments were applied in factorial combination to 2 × 2-m subplots within each of the treatments applied within one of the replicate blocks at each site; these subsidiary treatments were (i) litter removal by raking and (ii) addition of Calluna seed. The seed was added by applying cut Calluna shoots carrying abundant seed capsules at 100 kg shoots ha−1.

sampling methods

Within each treatment plot, the following measurements were made in the summer of each year from 1995 to 2000: (i) vegetation height, measured at 20 random locations with a sward stick (Stewart, Bourn & Thomas 2001), and (ii) cover of each species of higher plants estimated visually in four randomly positioned 1 × 1-m quadrats. Some bryophytes were assessed as composite groups. The cover of some species (Molinia caerulea, Eriophorum vaginatum, Empetrum nigrum, Calluna vulgaris) was assessed as both living and dead components. Within the subsidiary experiment, Calluna seedling emergence was counted in three randomly positioned 0·5 × 0·5-m quadrats in each of the treated plots from 1996 to 2000.

data analysis

Univariate analysis of individual responses

Univariate analyses of variance on untransformed and transformed data [square root for vegetation height and Calluna seedling counts, arcsine (√x (%)/100) for species cover] were carried out. A repeated-measured anova (proc glm; SAS 1989) was used to test the effects of region, moorland type, burning, grazing and herbicide and their interactions on vegetation species and height using a split-split-split-split plot design, with experimental site (region × moor type) as main plots > burning as subplots > grazing as subsubplots and herbicides as subsubsubplots. All analyses were done using transformed data; transformed means are presented for significant effects with a LSD (P < 0·05) computed using the appropriate error term. In the subsidiary experiment, the sites were used as the blocks in a randomized block analysis of variance.

Analysis of community response

Analyses of community-level responses were done using a multivariate analysis with a framework of constrained ordinations. Initially the entire data set (6 years × four experimental sites) was analysed using a detrended correspondence analysis (DCA) using a canoco for Windows (version 4.02) package (Ter Braak & Šmilauer 1998). In this analysis, and all subsequent multivariate analyses, the species data were transformed (ln x + 1) and there was no downweighting of rare species. The gradient length in this DCA analysis was 3·0 SD and subsequent analyses were done using canonical correspondence analysis (CCA; Ter Braak & Šmilauer 1998). Multivariate analysis of variance was then performed using the method developed by Ter Braak & Šmilauer (1998) using canoco for Windows. With this approach, the significance of each factor and interaction was determined using separate runs of CCA. In each run, the factors and their interactions were included as environmental variables, with experimental blocks included as covariables. Significance was assessed using restricted Monte Carlo permutation tests, with 499 permutations, using the methods of Ter Braak & Smilauer (1998). The model used was a split-split-split-split plot design, with region and moorland type at the top level, followed by burning, grazing, herbicide and time. Finally, the models underlying the analyses reported in detail were retested with 9999 permutations, as recommended by Legendre & Legendre (1998).


effects of moorland management treatments on individual vegetation measures

Vegetation height

Of all the treatments, herbicide application had the greatest and most consistent significant effect on vegetation height; there were no significant interactions with region, moorland type or other applied treatment. Where glyphosate had been applied, height was reduced significantly in 1996, the year after spraying (Fig. 1a), and this significant reduction was maintained throughout, although by 1999 and 2000, 5 years after treatment, the height in herbicide-treated plots was recovering and approaching the unsprayed control values. The higher rate of herbicide produced a greater reduction in height than the low rate (Fig. 1a) but this was not significant. Time was significant (P < 0·001), as was its interaction with herbicide application (P < 0·001). Few other treatments produced consistent significant differences. Burning was an exception, and showed a similar response at two sites (Peaks grey and Dales white). Immediately after burning, vegetation height was lower on burned sites compared with the unburned treatments (Fig. 2). There was rapid recovery in the following 2 years (1996 and 1997), with no significant difference between the two burning treatments, but in 1998 and 1999 the vegetation was taller in the burned sites than the unburned ones. By 2000 the treatments had converged (Fig. 2). Grazing had negligible significant effects on sward height over the 6-year period of the study, but the tallest vegetation was found on plots where grazing was excluded.

Figure 1.

Effects of glyphosate application on (a) vegetation height and (b) live Molinia cover in moorland dominated by Molinia caerulea between 1995 (year of application) and 2000. As there was no significant interaction between herbicide treatment and region the data shown are pooled across all four experimental sites (two regions × two moors types). Herbicide treatments: closed symbols, unsprayed control; open symbols, sprayed; dashed line, 0·27 kg ai ha−1; solid line, 0·54 kg ai ha−1. Pooled treatment means (n = 48) are presented with vertical bars representing the LSD (P < 0·05) and significance is denoted: n=P > 0·05, *P < 0·05, **P < 0·01, ***P < 0·001.

Figure 2.

Effects of burning on two moorlands dominated by Molinia caerulea between 1995 (year of application) and 2000: (a) Peaks, grey; (b) Dales, white. Burning treatments: closed symbols, unburned; open symbols, burned. Pooled treatment means (n = 18) are presented with vertical bars representing the LSD (P < 0·05) and significance is denoted: n=P > 0·05, *P < 0·05, **P < 0·01.

Effects of moorland management treatments on live and dead Molinia

Like vegetation height, the cover of live Molinia was reduced significantly only by herbicide application; again there were no significant interactions with region, moorland type or other applied treatment. Time was significant (P < 0·001), as was its interaction with herbicide application (P < 0·001). There was an immediate reduction in cover (from c. 40–45% to < 20%) in the year after spraying, which was followed by gradual recovery (Fig. 1b). Again by 2000, the high rate of herbicide application reduced Molinia cover to a greater extent but this was not significant. By 2000 Molinia was recovering but there was still a significant reduction compared with unsprayed controls.

Effects of moorland management treatments on the cover of non-target moorland species

The main treatment that produced consistent significant responses in the cover of individual species was herbicide application. Although other treatments yielded occasional significant differences, they were usually only detected once within the 6 years at one site, and accordingly are not discussed further here. Moreover, the variability in the data set was high, with the same species often exhibiting conflicting responses either between sites or between years. Those species that showed significant responses to herbicide at each site in at least one year were grouped into three classes: (i) those where cover was reduced in herbicide-treated plots (Calluna vulgaris, Empetrum nigrum, Vaccinium myrtillus); (ii) those where cover was increased in herbicide-treated plots (bryophytes, Erica tetralix, Juncus squarrosus, Polytrichum commune, sphagnum spp.); and (iii) those where conflicting results occurred, with increases and decreases at different sites/times (Anthoxanthum odouratum, Deschampsia flexuosa, Eriophorum angustifolium, Eriophorum vaginatum, Trichophorum cespitosum, Vaccinium oxycoccus). A typical example of a significant response from each group to herbicide application is shown in Fig. 3.

Figure 3.

Examples of significant species responses to glyphosate treatment. (a) Group 1: Calluna, cover reduced by herbicide; (b) group 2: Erica tetralix, cover increased by herbicide; (c, d) group 3: Deschampsia flexuosa, a species with conflicting responses at different sites or in different years. See text for further explanation. Herbicide treatments: control, unsprayed control; low, 0·27 kg ai ha−1; high, 0·54 kg ai ha−1 0. Pooled treatment means (n = 12) are presented with the LSD (P < 0·05) represented as a vertical bar.

Effects of litter removal and addition of Calluna seed on white sites

Calluna seedling density was significantly different in 1997 and 1999 (F = 98·4; P < 0·05) but not in the other years sampled, with the Peaks tending to have greater densities (mean values over the period: Dales = 2·9 seedlings m−2 compared with the Peaks = 11·1 seedling m−2). However, there was a large variation in response between treatments. The most significant main treatment effect was a positive response to herbicide application (F = 9·39, P < 0·05), with 8·2 and 9·4 seedlings m−2 found in the low and high application rates, compared with 1·6 m−2 where no herbicide was applied. There were also consistent significant effects of litter removal, Calluna seed application and their interaction (raking F > 18·0, P < 0·001 in every year except 2000, where F= 3·9 and was not significant; seeding F > 19·0, P < 0·001 in every year except 2000, where F= 6·9, P < 0·01; raking × seeding F > 24·0, P < 0·001 in every year except 2000, where F= 8·4, P < 0·01). Repeated-measures analysis confirmed that there were significant changes during the time period with respect to litter raking (F = 13·99, P < 0·001), seeding (F = 4·99, P < 0·01) and their interaction (F = 44·18, P < 0·001).

The most important result was the large number of Calluna seedlings that established in 1996, the year after the seed was applied. At this time, the most successful treatment was where seeds were applied without litter removal (24 seedling m−2) followed by litter removal (2·5 seeds m−2). Surprisingly, no seedlings were found where litter was removed and seeds were applied. No seedlings were found in the unseeded, unraked plots with no seed addition throughout (Fig. 4). Over the next 4 years the densities declined in the seeded plots until they reached 0·2 seedlings m−2 in 2000, similar densities to those found in plots with no seeds applied but with litter removal. However, where seed was applied and litter removed there was a slight increase to 0·8 seedlings m−2 in 2000, 6 years after treatment (Fig. 4).

Figure 4.

Effects of removal of litter by raking and application of Calluna seed on Calluna seedling establishment on white moorland dominated by Molinia caerulea between 1996 (year of treatment) and 2000. Litter removal treatments: dashed line, no removal; solid line, litter removed; seeding treatments: open symbols, unseeded; closed symbols, seeded. Pooled treatment means (n = 36) are presented with vertical bars representing the LSD (P < 0·05) and significance is denoted: n=P > 0·05, **P < 0·01, ***P ≤ 0·001.

effects of moorland management treatments on community response

The multivariate analysis of variance showed that all but one of the treatments and their interactions were significant (Table 2), indicating that they were all influencing the vegetation in some way. These responses were complex, and not all of the interactions were easy to explain. Each line in the analysis of variance table (Table 2) was derived from a separate constrained CCA run, and each had its own biplot displaying species and environmental factors (treatment–interactions). We concentrated on three results from this analysis: (i) the interaction between all experimental factors, ignoring time (Fig. 4); (ii) the effect of time on its own (Fig. 5); and (iii) the interaction between all factors including time (Fig. 6). The aim was to identify broad trends associated with each of the main factors. In each of these graphs the species biplot is presented and then the treatment interaction positions are plotted separately in the same space; figure quadrants are denoted L/U, lower/upper, and L/R, left/right. The models produced in all three analyses were significant when tested with 9999 permutations: (i) trace = 0·533, F= 5·82, P < 0·0001; (ii) trace = 0·157, F= 23·52, P < 0·0001; (iii) trace = 0·971, F= 1·68, P < 0·0001. In the graphical description of the biplots in analyses 1 and 3, the two levels of grazing and herbicide application rates have been pooled for clarity.

Table 2.  Results of the multivariate analysis of variance using restricted permutation tests (499 permutations) of an experiment carried out at two moor types in two regions where burning, grazing and herbicide treatments were applied to control Molinia and enhance moorland vegetation over a 6-year period. Np, number of permutable units; Trace (Tr) = total sums of squares; F, F-ratio; P, P-value
FactorAloneInteraction with time (Np = 864)
Region 1  40·197110·90·002 50·39422·90·002
Moor 3  80·103 54·50·052150·20816·10·002
Region × moor type 3  80·365 76·60·002150·26410·70·002
Burn 1 160·006  3·80·038 50·16811·50·002
Region × burn 1 160·048 10·80·002 50·167 7·40·002
Moor type × burn 3 160·011  3·70·004150·230 8·20·002
Region × moor type × burn 3 160·018  3·00·004150·303 5·60·002
Graze 2 480·017  5·70·002100·188 8·40·002
Region × graze 2 480·029  3·20·016100·161 3·20·002
Moor type × graze 6 480·031  5·20·002300·264 6·20·002
Burn × graze 2 480·029  3·80·004100·213 4·30·002
Region × moor type × graze 6 480·046  3·90·002300·355 4·40·002
region × burn × graze 6 480·046  3·10·002100·296 3·30·002
Moor type × burn × graze 6 480·053  3·60·002300·315 3·60·002
Region × moor type × burn × graze 6 480·083  2·90·002300·412 2·90·002
Herb 21440·039 13·20·002100·21910·00·002
Region × herb 21440·210 26·80·002100·301 6·20·002
Moor type × herb 61440·145 17·60·002300·564 6·60·002
Burn × herb 21440·047  5·30·006100·134 2·30·002
Graze × herb 41440·062  4·40·012200·169 1·90·002
Region × moor type × herb 61440·390 25·90·002300·564 6·60·002
Region × burn × herb 21440·22713·20·002100·344 3·30·002
Region × graze × herb 41440·251 9·60·002200·402 2·50·002
Moor type × burn × herb 61440·166 9·30·002300·326 3·00·002
Moor type × graze × herb121440·188 6·80·002600·381 2·30·002
Burn × graze × herb 41440·080 2·90·530200·228 1·20·002
Region × moor type × burn × herb 61440·42213·60·002300·649 3·70·002
Region × moor type × graze × herb121440·458 9·90·002600·739 2·80·002
Region × burn × graze × herb121440·289 5·60·002600·515 1·60·002
Moor type × burn × graze × herb 41440·231 4·20·002200·964 1·70·002
Region × moor type × burn × graze × herb121440·533 5·80·002600·971 1·70·002
Time 58940·15723·50·002
Figure 5.

Biplots from the CCA analysis of variance testing the interactions between region, moorland type, burning, grazing and herbicide. (a) The species biplot; (b) distribution of the four sites (regions × moorland types: squares, Peaks white; triangles, Peaks grey; circles, Dales white; stars, Dales grey); (c) the burning treatments at the four sites (burning, open; no burning, closed); (d) the grazing treatments at the four sites (grazing, open; no grazing, closed); (e) the herbicide treatments at the four sites (herbicides, open; no herbicide, closed). Species key: Ao, Anthoxanthum odouratum; As, Agrostis spp.; BG, bare ground; Ca, Chamenerion angustifoilium; Ce, Carex echinata; Csp, Carex spp.; Cv, Calluna vulgaris; Dc, Deschampsia cespitosa; Df, Deschampsia flexuosa; Dr, Drosera rotundifolia; Ds, Dryopteris spp.; Ea, Eriophorum angustifolium; En, Empetrum nigrum; Et, Erica tetralix; Ev, Eriophorum vaginatum; Fo, Festuca ovina; Gs, Galium saxatile; Hl, Holcus lanatus; Ji, Juncus inflexus; Js, Juncus squarrosus; Jsp, Juncus spp.; Lc, Luzula campestris; Ls, lichen spp.; Mc, Molinia caerulea; Ms, moss spp.; No, Narthecium ossifragum; Ns, Nardus stricta; Pc, Polytrichum commune; Pe, Potentilla erecta; Ps, Pedicularis sylvatica; Ss, sphagnum spp.; Vm, Vaccinium myrtillus; Vo, Vaccinium oxycoccus; Tc, Trichophorum cespitosum.

Figure 6.

Biplot from the CCA analysis of variance testing the effects of time alone in a study assessing the effects of region, moorland type, burning, grazing and herbicide. The species biplot is presented with the positions of each year (1995 = t1 to 2000 = t6). Species key follows Fig. 5.

Analysis 1

The species biplot (Fig. 5a) showed a distribution of species away from the Molinia-dominated vegetation, with few species (UR) in a negative direction on axis 1, which was associated with species typical of acidic grassland (Anthoxanthum, odouratum, Galium saxatile, Festuca ovina), and in a negative direction on axis 2, towards dwarf shrub species typical of moorland (LR; Calluna, Erica tetralix, Empetrum nigrum) and wetter vegetation (LL, Deschampsia cespitosa, Narthecium ossifragum) and nutrient-poor acid grassland (Nardus stricta).

The distribution of the treatment interactions within this species space showed obvious regional and moor differences in both the area of the species graph occupied and the variation within each site (Fig. 5b). Thus, the Peaks were located within the Molinia moorland area, with the white site being clearly distinct from the grey one. However, both Dales sites were located within the more species-rich grassland area, and the separation between white and grey was much less distinct, although the grey sites tended to be closer to the bog vegetation.

The effects of the different treatments showed that there was a great deal of overlap at most sites between the treated vs. untreated, but there were indications of different patterns at each site. For burning at both Peaks sites there was little difference between burned and unburned, although at the white site the treatments nearest the dwarf shrub vegetation had all been burned (Fig. 5c). At the Dales sites there was little obvious effect of burning. Grazing had little impact at both white sites but at the grey sites the grazed treatments tended to be located near the bog and moor vegetation (LL, LR; Fig. 5d). Herbicide had little effect on community composition at the Peaks sites, but at both Dales sites there was a tendency for the herbicide-treated plots to be positioned closest to the acid grassland species (Fig. 5e).

Analysis 2

Molinia was plotted almost at the origin of the species biplot, with the majority of species separated along axis 1 and the separation of a few species on axis 2 [Deschampsia cespitosa and lichens at the positive end (UL) and Carex echinata, Juncus squarrosus, Juncus spp., Empetrum nigrum and bare ground at the negative end (LL); Fig. 6]. The individual years were plotted around the periphery. Years 1, 2 and 3 were plotted close together at the positive end of axis 1, followed by a move to the negative end of both axes for years 4 and 5, then in year 6 a move to the positive end of axis 2. This implied that species change is slow in these communities and treatment effects might be delayed several years after implementation.

Analysis 3

The species biplot (Fig. 7a) showed Molinia almost at the origin, with a gradient on axis 1 to the negative end towards acid grassland and a gradient on axis 2 between Calluna and Erica tetralix in the UR and bog species and Empetrum nigrum in the LR. The treatments at the different sites showed quite different responses through time; the white sites were more clustered and showed less obvious change through time (Fig. 7b,d). This result should be expected. Moreover, as in analysis 1, the sites occupied different parts of the biplot, Peaks being closer to Empetrum nigrum and Dales being closer to acid grassland. The grey sites, in contrast, showed a much greater spread and change through time (Fig. 7c,e). The Peaks site was positioned in the UR quadrant at the start, with a gradual move through years 4, 5 and 6 to have a greater composition of Empetrum nigrum. The Dales grey site showed a similar but more gradual change in the direction of acid grassland.

Figure 7.

Biplots from the CCA analysis of variance testing the interactions between region, moorland type, burning, grazing, herbicide through time (1995–2000). (a) The species biplot; (b) Peaks white; (c) Peaks grey; (d) Dales white; (e) Dales grey. The generalized trend is indicated with the arrow. The species key follows Fig. 5.


The aim of this experimental multi-site study, in two regions of upland England with different climates and management approaches, was to help develop a national strategy for controlling Molinia and increase the amount of dwarf shrubs within the vegetation. Originally, the study included a pair of experiments in Exmoor, southern England, but these were abandoned after 3 years because of persistent vandalism (Todd et al. 2000). The study included, within each region, two different types of Molinia-dominated ecosystem: white moorland, which was almost a Molinia monoculture, and grey moorland, where there was a mixture of Calluna and Molinia.

Multi-site studies with multifactorial treatments are inevitably complex (Pywell et al. 2002) and difficult to analyse. We used a combination of univariate and multivariate analysis. Both were important in helping to understand the complex relationships within these data. The univariate analyses identified that there were important single-treatment effects, especially the effects of herbicide, seed addition and disturbance, and the multivariate analyses identified more subtle effects associated with community change across the different sites.

effects of the applied management treatments: univariate analyses

Effects on Molinia

The preliminary results from this study show that Molinia height and cover can be reduced successfully, although it was beginning to recover within 5 years. It was, however, very difficult to derive management prescriptions that will work in all regions. The one exception was that Molinia cover was always reduced by the application of the herbicide glyphosate. Of the other treatments, the effects were either site-specific or were only found on some sampling occasions.

Spring burning, for example, removed the leaf litter accumulated from previous years and reduced vegetation height temporarily, but there was rapid Molinia recovery, eventually to levels greater than the untreated plots. Todd et al. (2000) suggested that compensatory growth after burning was a possible reason for the high Molinia cover found in their Exmoor study. Ross, Adamson & Mood (2003) suggested that burning when combined with low grazing increased Molinia dominance, which is consistent with Grant, Hunter & Cross (1963), who argued that burning Molinia was counterproductive unless it was closely grazed subsequently.

The effects of the grazing treatments in this study were inconsistent between regions. Hard summer grazing is known to control Molinia (Grant, Hunter & Cross 1963; Hulme et al. 2002) and maintain dwarf shrubs. However, limited effects of grazing were found here, partly because the ESA and CS scheme grazing pressures are relatively low and were implemented over a relatively short period (6 years). Grazing may have been more effective if (i) a greater pressure had been used, (ii) cattle or ponies had been included or (iii) the experiment had been carried out over a much longer period. However, these suggestions need to be tested further.

Application of glyphosate reduced the height and cover of the target species (Molinia) at all sites, and the effect of the low rate (0·27 kg ai ha−1) was as effective as the high rate (0·54 kg ai ha−1), although after 3 years the Molinia regrowth was greater where the low rate had been applied. There is no need to use the higher rate of glyphosate for initial control, but it may have advantages in maintaining control for longer. A better strategy might be to repeat the low application rate after c. 4 years. The use of repeat applications, albeit of selective herbicides, has been used successfully in large-scale Molinia control schemes (G. Eyre, personal communication).

The use of herbicide on white sites should not cause much damage to non-target heathland species, because there are usually very few other species found on such sites. However, on the grey sites glyphosate application affected the Calluna cover adversely, decreasing the amount of live Calluna and increasing the amount of dead Calluna. Glyphosate is a total herbicide and is recommended for Calluna control in forestry (Tabbush 1984). Tabbush (1984) recorded significant Calluna death where glyphosate was applied at 1·8 kg ai ha−1 (three times the maximum rates used here), but Marrs (1984) has shown rapid recolonization from seed (within 2 years) after Calluna was completely killed by a total herbicide. Thus glyphosate must be used carefully in situations where Calluna and Molinia coexist. The ideal solution would be to use a selective herbicide (graminicide) that does not damage or kill Calluna but kills Molinia. A range of graminicides has been tested (King & Davies 1963; Phillips 1995; Milligan 1998; Milligan, Putwain & Marrs 1999, 2003; Ross, Adamson & Mood 2003) with varying degrees of success. Unfortunately, these graminicides are not presently approved for Molinia control in the British uplands. Some other species were affected by glyphosate, but the results were highly variable between sites and years. Nevertheless, three groups of responses were tentatively identified: those species adversely affected by herbicide (Calluna vulgaris, Empetrum nigrum, Vaccinium myrtillus), those species that could apparently respond positively to herbicide (Erica tetralix, Juncus squarrosus, bryophyte cover plus the Polytrichum commune and sphagnum components) and those with variable responses to herbicide (Anthoxanthum odouratum, Deschampsia flexuosa, Eriophorum angustifolium, Eriophorum vaginatum, Trichophorum cespitosum, Vaccinium oxycoccus).

Although Molinia was beginning to recover towards the end of the 6-year period, the management objective should not be to eradicate it. Grant et al. (1976) have shown that Molinia, along with sedges (Carex spp.), Trichophorum cespitosum and other grasses (Deschampsia flexuosa, Agrostis canina), provides important forage for grazing animals in summer. As the digestibility of these species declines in autumn, choice of forage is progressively focused more towards Calluna and Vaccinium myrtillus, and then Eriophorum vaginatum in late winter. Essentially herbicide application helps reduce Molinia cover and gives an opportunity for other species to establish.

Restoration of Calluna on Molinia-dominated white moorland

The major factors likely to impede the recolonization by dwarf shrub species are the amounts of Calluna seed remaining in the seed bank and the depth of Molinia leaf litter, which can inhibit the germination of any Calluna seedlings (Miles 1979; Putwain, Gillham & Holliday 1982). Assessment of the numbers of viable seed in seed bank samples of the white moorland sites showed no Calluna seed present, and hence seed would have to be added to accelerate dwarf shrub reestablishment (P.A. Todd, unpublished data). However, the cover of dead Molinia and litter depth was reduced in 1997, especially where herbicide was applied (Todd et al. 2000). This indicates that herbicide either reduced litter production or increased litter removal through decomposition or wind blow. Irrespective of these factors, it implies that conditions were becoming more suitable for Calluna establishment from seed. However, the results from the subsidiary disturbance study indicated that further intervention was needed, either physical disturbance/removal of leaf litter or the addition of Calluna seed. This has been confirmed in more extensive field experiments (Milligan 1998; Milligan, Putwain & Marrs 2003) and in practical land management projects (G. Eyre, personal communication).

The subsidiary study also showed that the greatest Calluna seedling densities were found immediately after treatment in the most successful treatment combinations (litter removal +Calluna seeds) in white Molinia stands. Moreover, there were important differences between the two regions, with Peaks having greater densities than the Dales. The low seedling establishment found in the Dales may be related to a combination of interacting factors, including a microtopography with an abundance of bryophytes in tussock hollows, which restricts the area suitable for Calluna germination, and waterlogging caused by a high water table at this site. Calluna seedlings are known to be susceptible to weather fluctuations (Miller & Cummins 1987) and they need a high soil and atmospheric moisture concentration throughout the establishment period, but they are intolerant of waterlogging (Bannister 1964; Gimingham 1972).

Even though Calluna established in the year after the seed was added, in the second year a large reduction in established seedlings was found. We do not know the reason for this mortality but suggest a range of potential factors, including adverse weather (notably waterlogging, frost heave and freezing winds), possibly coupled with trampling or browsing by grazing livestock. If high grazing pressures are implicated at this sensitive point in Calluna recovery, grazing pressures may have to be reduced still further during the initial phases of Molinia control/Calluna restoration.

regional differences in community response

A major finding of this study was the clear separation of the four experimental sites in the multivariate analysis of community composition. The sites were selected on crude visual assessments of their dominant species: white being visually mono-dominant Molinia and grey being mainly Molinia but with obvious mixtures of dwarf shrubs. However, when detailed analysis of the floristic composition based on species cover was done, regional–site treatment interactions were one of the most important factors separating responses. While all sites had a high Molinia cover and this reduced and recovered during the experimental period, the overall change in species composition differed between sites. The Peaks white site was predominantly Molinia and remained so, the Peaks grey was distinct from its white counterpart and had a dwarf shrub moorland component. In the Dales, the distinction between white and grey was more blurred, and there was a more pronounced acidic grassland component. These results indicate that the site starting point is important, and even when apparently similar sites are selected either slight nuances in species composition are present or differences in site factors (soils, microclimate, etc.), and these differences may significantly influence the subsequent post-restoration successional trajectory.

The multivariate analysis also showed that most applied treatments had a greater effect when community composition was considered. This is to be expected because here species diversity is low, and most species are present at small cover and at variable densities within the Molinia and MoliniaCalluna matrix. The main conclusions to come from this study were that (i) burning had little effect, (ii) grazing had little effect on white sites but a positive effect towards moorland–bog development on grey sites, and (iii) while herbicide reduced species cover, the overall effect on community composition over the entire experimental period showed little effect at the Peaks sites but a positive development of acid grassland species at that Dales sites. These results emphasize that different and interacting responses occurred in the different sites.

The temporal analysis also indicated that community response was non-linear, with sudden jumps between years, which might have been a result of treatment-induced disturbance providing niches for colonization. This implies that where restoration treatments are applied to Molinia-dominated moorland, effects may persist well beyond the time of application, and this must be considered in any monitoring scheme. The shift in both grey sites towards an acidic grassland composition is of concern, given that the main objective for these sites is to retain and enhance a CallunaMolinia mixed moorland.

implications for moorland management

Moorland degraded by an increase in Molinia cover is clearly affected by regional differences, brought about by past history and management and present management prescriptions allowed under agri-environment schemes or conservation restoration schemes. In order to restore such degraded sites to more diverse communities, in keeping with biodiversity action plans (Anonymous 1995a,b), there are several important steps to take. First, it is necessary to determine the scale of the degradation. If the vegetation is predominantly a Molinia-dominated white site, then a complete restoration package will be necessary, which will involve initial control of the Molinia, disturbance of the litter layer and the addition of seed (Milligan 1998; Todd et al. 2000; Milligan, Putwain & Marrs 2003). If it is a mixed stand with other species present, then the aim should be to maintain the mixture and prevent further degradation. Second, it is wise to determine the species composition of the vegetation in detail at the outset, as there is marked variability between sites, even when the overall appearance looks similar. Determining this will influence the choice of target endpoint and the methods needed to achieve it. Third, accurate monitoring is needed for some time after the treatments have been applied, because impacts can continue to occur for at least 6 years. These systems change relatively slowly because of the adverse climatic conditions, and a long-term strategy is required. Fourth, it may be wise to consider a range of communities as potential endpoints, because here the vegetation change in the grey sites implied a move towards acidic grassland rather than the preferred dwarf shrub community. While it is possible that the acidic grassland is a stage on the way towards an increase in dwarf shrub, this cannot be guaranteed, so a decision would have to be taken to either (i) accept an acidic grassland as an alternative endpoint or (ii) apply additional restoration treatments to encourage dwarf shrub invasion. This might involve the application of graminicides (Milligan, Putwain & Marrs 1999, 2003) or the use of herbicide (glyphosate or selective graminicide) applied by weed wiper (Milligan 1998). Finally, once a mixed MoliniaCalluna mosaic has been achieved, it is essential to manage the moorland using a balanced programme of appropriate burning regime plus controlled grazing to prevent further damage to, or loss of Calluna, and thus ensure long-term sustainability (Hulme et al. 2002). Further work is needed to develop such management regimes for a range of grass–Calluna mosaics. However, as moorlands are likely to require management for multiple uses (farming, amenity, tourism, sporting, water collection, etc.), the exact prescriptions will vary depending on the end-use required.


We thank the moor owners, farmers and gamekeepers who gave permission to work on their land and assisted us in the study. Monsanto plc kindly provided the Roundup, and Penny Anderson Associates are thanked for assistance with field recording in 2000. Defra funded this work.