Manipulation of nutrients and grazing levels on heather moorland: changes in Calluna dominance and consequences for community composition

Summary

• 1
  Experimental studies of the combined effects of herbivory and the availability of nutrients on semi-natural communities remain relatively scarce. Here we report the effects of 6 years of nutrient addition (N, P and K) and protection from grazing on moorland plant communities in the Scottish uplands, particularly on the cover of the dominant Calluna vulgaris. We also recorded the cover of vascular plants and bryophytes, to assess the impact of changes in Calluna dominance on competing species.
• 2
  Grazing in combination with nitrogen addition caused the greatest decline in Calluna cover, typically 40–50%, but nitrogen addition did not cause a significant decline in Calluna on plots protected from grazing. More Calluna shoots were browsed on nitrogen-treated plots than on unfertilized ones, presumably because grazing animals preferred fertilized Calluna.
• 3
  Nitrogen addition allowed grasses to increase in cover, especially on grazed plots. However, Nardus stricta, Festuca ovina and Agrostis sp. all declined in fenced areas but increased in grazed plots, whereas Deschampsia flexuosa and Festuca rubra increased in fenced plots.
• 4
  The effects of grazing and nutrient addition varied markedly between sites, possibly because of differences in soil moisture and organic matter. Nitrogen addition, however, increased soil nitrogen mineralization rates in both glens.
• 5
  Fencing increased the cover of grazing-intolerant plants with low nutrient demands (as classified by Ellenberg and suited species scores) that were categorized as competitive plants by Grime's CSR model. Plots receiving nitrogen and phosphorus had more nutrient-demanding plants able to tolerate high grazing pressure that were often classified as ruderals.
• 6
  The impact of nitrogen addition on the cover of Calluna and on competing grass species in the community critically depends on the level of grazing. Changes in community composition caused by grazing and fertilizer addition can be explained in terms of the ecological tolerances of individual species, allowing us to predict the types of plants that are likely to increase or decrease in cover.

Introduction

Herbivory is one of the principal drivers of vegetation change. Grazing animals affect the growth and reproduction of individual plants, but their selective grazing also alters the competitive balance between species and so may affect community structure (Crawley 1997; Augustine & MacNaughton 1998). However, other environmental factors constrain the structure of plant communities, principally the availability of resources (nutrients, light and water) and the ability of different species within the communities to acquire these resources. The outcome of competition for limiting resources is likely to be modified by herbivores: grazing may change a plant's ability to acquire resources (e.g. by altering its allocation to above- vs. below-ground parts, Hartley & Amos 1999), or by damaging the performance of potential competitors (Louda et al. 1990). Although there are now several theoretical models for the interactions between herbivory and competition for nutrients (Holt et al. 1994; Grover 1995) or light (Huisman & Olff 1998; Huisman et al. 1999), studies that evaluate the interactions between herbivory and resource competition experimentally in the field still remain scarce. Even fewer of these are relatively long-term studies in semi-natural communities (but see Turkington et al. 1993; Van der Wal et al. 2000; Van der Wal et al. 2003). A better understanding of these interactions in the field will increase knowledge of the mechanisms by which one plant species replaces another in grazed ecosystems.

Upland heather moorland dominated by Calluna vulgaris (L. (Hull)) is one of the most distinctive ecological habitats in Europe, with high value for conservation, tourism and game management (Usher & Thompson 1993; Thompson et al. 1995). Grazing has a major influence on the vegetation in this habitat: over-grazing by sheep and deer is one factor thought to have contributed to the replacement of heather-dominated moorland by rough grassland in recent years (Welch 1984b; Welch 1986; Welch & Scott 1995). There have been several attempts to establish general ‘rules’ regarding grazing impacts on moorlands (Welch 1984a,b; Grant et al. 1978), but these mask a large degree of complexity. For example, the impacts of grazing are known to be highly spatially variable (Hester & Baillie 1998; Palmer et al. 2003) and the type and age of herbivore present can also affect grazing impacts, both in terms of off-take rates and in terms of their preferences for certain vegetation types (Clarke et al. 1995; Armstrong et al. 1997).

The effects of grazing on moorland vegetation are also likely to be modified by site-based factors, such as altitude, aspect and, most importantly, soil conditions. Soil nutrient levels play an important role in the competitive balance between Calluna and grasses, such that losses of Calluna cover under heavy grazing appear to be more rapid on more base-rich soils (Hartley 1997). In the Netherlands, recent losses of Calluna cover have largely been attributed to increased atmospheric deposition of nitrogen, rather than to grazing (Aerts & Berendse 1988; Bobbink et al. 1998). Calluna is a relatively slow-growing plant adapted to nutrient-poor environments. Nitrogen addition shifts the competitive balance in favour of the faster growing grasses, which are better able to acquire the extra nutrients than woody species (Heil & Bruggink 1987; Aerts et al. 1990; Hartley & Amos 1999). Although the impact of increased nitrogen deposition has been the focus of much research on Calluna-dominated communities, both in the Netherlands and in the UK (e.g. Pitcairn et al. 1995), the impacts of other nutrients are less well studied. However, previous research suggests that the effect of nitrogen addition on Calluna performance will be modified by phosphorus availability (Whitehead et al. 1997; Kirkham 2001).

Both grazing and nutrient inputs can alter the competitive balance between Calluna and invading grass species (Berendse 1985), but the combined effects of these factors on plant communities is less well known. Increased atmospheric nutrient deposition is likely to drive the community composition towards one dominated by plants with high nutrient requirements, while increases in grazing pressure are likely to result in an increase in the cover of plants that are tolerant to grazing. Hence when both drivers change there may be an interacting effect on the plant community (Hartley 1997; Van der Wal et al. 2003). Increased nutrient availability may result in more nutrient-rich plants, which will attract more grazers (Duncan et al. 1994), increasing the grazing pressure, which will lead to a community dominated by grazing-tolerant plant species. The increase in grazers will also result in higher nutrient levels from dung and urine, and more disturbance from trampling by grazing animals.

In an upland Calluna-dominated moorland system, we carried out an experimental manipulation of grazing (with fencing) and nutrient levels (using fertiliser), and measured the effects on Calluna cover and the species composition of the moorland community. Our general aim was to quantify the interacting impacts of grazing and soil nutrient addition on rates of vegetation change in moorland systems. Specifically, we tested whether: (i) loss of Calluna cover would be most rapid on the grazed plots that had also received fertiliser, as these were the plots where herbivory would be most intense; (ii) nutrient addition would increase grass cover, but which grass species successfully invaded the Calluna canopy would depend on grazing intensity; (iii) grazing and nutrient addition would have significant and interacting effects on the species composition of the community as a whole; and (iv) which particular species increased and which decreased in response to the nutrient and grazing treatments could be explained in terms of the ecological tolerances of individual species.

Methods

study sites

The study was conducted on two upland moors in the Grampian mountains in north-east Scotland: Glen Clunie (500 m a.s.l., national grid reference NO 139815) and Glen Shee (500 m a.s.l., national grid reference NO 142812). Both glens have an average rainfall of c. 890 mm. At Glen Clunie the soil was a wet, acidic peaty podzol overlying a base-poor schist (Welch 1984a; Welch & Scott 1995) and the grazing pressure from red deer and sheep was moderately high (40–50% of Calluna current year's shoots browsed; Hartley 1997; Gardner et al. 1997). This glen has an area of around 6500 hectares, which is grazed by about 750 deer and, in the summer only, about 1800 sheep. The soils at Glen Shee were drier and were of the brown earth type overlying dolmitic outcrops; grazing pressure was higher (over 60% of current year's shoots browsed), particularly by sheep. The glen area supports around 250 deer and 3000 sheep all year round in an area of only 4400 hectares, of which much is higher hill ground, hence the sheep concentrate their grazing on the valley floor. On both moors, the vegetation consisted of a fine mosaic of Calluna vulgaris, other ericoids and graminoid species and was identified as National Vegetation Classification H12 (Calluna vulgaris-Vaccinium myrtillus heath) (Rodwell 1991); the vegetation in Glen Shee had a greater abundance and variety of herbs. The estimated annual nitrogen deposition at the sites is 18 kg N ha⁻¹ year⁻¹ (M.A. Sutton, personal communication).

experimental design

On each moor, two study sites were established, sites C1 and C2 in Glen Clunie and sites S1 and S2 in Glen Shee. At each of the four sites, four blocks of vegetation were selected and two of these were fenced in March 1993 to exclude red deer, domestic sheep and mountain hares. Each block contained four 5 × 3 m experimental plots to which a range of nitrogen, phosphorus and potassium fertilizer treatments were applied. Thus there were 16 blocks and a total of 64 plots. The design was a fractionated 2³ factorial: four of the eight possible treatment combinations (N, P, K, NPK) were applied to the plots in one fenced and one unfenced block at each site and the other four treatments (NP, KN, KP, no fertiliser) were applied to plots in the second pair of fenced and unfenced blocks. Nitrogen was added as ammonium nitrate at 75 kg N ha⁻¹ year⁻¹, phosphorus was added as ‘superphosphate’ at 12.5 kg P ha⁻¹ year⁻¹ and potassium as potassium sulphate at 25 kg K ha⁻¹ year⁻¹. The fertiliser was applied in solid form annually as one dose in spring and one in early summer, beginning in April 1993.

soil analysis

Two soil cores (5 cm diameter by 10 cm deep) were collected from each of the 64 plots in July 1997, 4 years after the start of the experiment. One core from each plot was weighed, oven-dried at 40 °C for 48 hours, and re-weighed to determine moisture content. The sample was then sieved through a 2 mm sieve. Organic matter content was estimated by loss on ignition at 550 °C (Allen 1989) on a 1 g sample of soil. The second core from each plot was split into two vertical halves and used to determine nitrogen mineralization rate. The first half was extracted in 1 m KCL, and nitrate and ammonium levels were determined following the method in Allen (1989). The second half was placed in a gas-permeable polythene bag and incubated at 10 °C for 4 weeks before being extracted in 1 m KCl and analysed for nitrate and ammonium as before. Mineralization rate was calculated from the difference in ammonium and nitrate concentrations between the two extractions and expressed on a gram dry weight basis (Emmett & Young 1998).

vegetation sampling

From each 3 × 5 m plot, three randomly chosen 1-m² subplots were selected on which all subsequent measurements were carried out. The values from the three subplots were combined to give a mean value for each plot. All analyses were conducted on the plot means. In May of each year a point quadrat (80 pins) was used to estimate the ground cover of Calluna and grasses in each subplot. In addition, each current year's Calluna shoot touched by a pin was recorded as grazed or ungrazed to give an estimate of annual grazing. The height of the Calluna canopy was recorded at two randomly selected points within each 1-m² subplot. Data for all the above parameters were recorded annually, but almost all results are presented as the change in a given parameter over the 6-year period from 1993 to 1999. The species composition of the whole community was recorded in July 1999. The vegetation cover (as percentage ground cover) for all species of vascular plants, bryophytes and lichens in each of the 1-m² subplots was estimated by eye to the nearest 5% as agreed by two observers. Nomenclature followed Stace (1991) for higher plants and British Bryological Society (2003) for bryophytes.

data analysis

Soil and vegetation parameters

The effects of protection from grazing and fertiliser addition on the changes (after 6 years) in cover and canopy height of Calluna, in the cover of grass, and on the mean annual browsing damage to Calluna, were analysed by anova using the GLM procedure in SASv8 (SAS 1999). The effects of the site and the fencing and fertiliser treatments on soil characteristics were also analysed by anova in the same way. Because the experimental design was fractionated, there were two levels to the analysis. The effects of site (3 d.f.), fencing (1 d.f.) and any interaction between them (1d.f.), were tested against a site × fence × block error term (8 d.f.). The effect of fertiliser (and any interaction between fertiliser and fencing) was tested against a site × block × plot error term (48 d.f.). Third order interactions (e.g. N × P × K) have no replicates in this design, so cannot be tested for.

Species composition

The vegetation cover data from 1999 were analysed by the constrained linear ordination technique redundancy analysis (RDA) using the computer programme canoco 4 (ter Braak & Šmilauer 1998). Species that occurred infrequently (< three times) were removed from the data set as rare species may have an unduly large influence on the analysis. The treatment variables used in the analysis were the grazing (fencing/no fencing) and fertiliser treatments (N, P, K). The four sites were included as covariables (i.e. concomitant variables whose effect must be partialled out before estimating the effect of the treatment variables). The vegetation data were transformed by a log_e (y + 1) transformation. The model was tested using a Monte Carlo permutation test, which was restricted for the split plot design of the experiment, following the method of Baar & ter Braak (1996).

A table showing the significance of each treatment was compiled on the basis of 15 different (partial) RDA analyses, following the method of Van Dobben et al. (1998). The significance of each treatment was tested by specifying the treatment to be tested as the only variable with the remaining treatments as covariables within the model. The model was then tested using a restricted Monte Carlo permutation test (9999 permutations). The effect of fencing was tested by using the option in canoco to restrict the permutation for a split design with whole plots being freely exchangeable and no permutation at the split plot level. The effects of each of the fertiliser treatments were tested in a similar way, using an unrestricted permutation test within whole plots, with the whole plots specified as covariables and as blocks (ter Braak & Šmilauer 1998). The interactions between fencing and fertilisers and between site and fencing were also tested.

Analysis of species characteristics

The realized ecological niches in which the vascular plants were able to grow in the different treatments were characterized using Ellenberg values (Ellenberg 1988), modified for British plants by Hill et al. (1999), and by suited species scores (Critchley & Smart 1996; Critchley 2000). Ellenberg scores for moisture and soil fertility (nitrogen) and suited species scores for grazing were used. The primary ecological strategies of plants based on their ability to survive in conditions of low stress and low disturbance (competitor species, C), low disturbance and high stress (stress-tolerator species, S) and high disturbance and low stress (ruderal species, R), were calculated using Grime's CSR scores (Grime 1977, 1979; Grime et al. 1995). For each plot, the mean score for each of the Ellenberg, suited species and Grime CSR indices was calculated and weighted by the square root of the average cover of the species on the three subplots within the plot, using the weighted option in the SAS means procedure (SAS 1999). The scores showed if the plots were dominated by vascular plants able to tolerate high levels of grazing, soil fertility or moisture and if the life-history strategy of the dominant plants was ruderal, stress-tolerator or competitor. The effects of site, fencing and fertiliser treatments on plot scores for Ellenberg, suited species and CSR scores were analysed by anova using the GLM procedure in SAS (as above). Differences between pairs of treatments were then tested by comparing least squared means.

Results

soil chemical properties

The soil characteristics varied between sites, hence site had a significant effect on soil moisture content (F_3,8 = 18.72, P < 0.001) and organic matter content (F_3,8 = 19.70, P < 0.001). Sites in Glen Shee (S1 and S2) had lower soil moisture (around 35%, P < 0.0001 for all contrasts) and lower organic matter (around 20%, P < 0.001 for all contrasts) than those in Glen Clunie (C1 and C2) (around 65% moisture and 70% organic content). Fertiliser treatments had no significant effect on soil moisture or organic matter. Nitrogen addition had a significant effect on net nitrification (F_1,48 = 6.47, P < 0.05) and ammonium accumulation (F_1,48 = 5.47, P < 0.05) (Fig. 1). Nitrogen-treated plots had higher rates of net nitrification and ammonium accumulation and, as a consequence of this, the net mineralization was also higher on nitrogen-treated plots than on untreated ones (F_1,48 = 7.37, P < 0.05).

Figure 1. Effect of site and nitrogen addition, net nitrification and ammonium accumulation (mean ± 1 SE are shown, n = 8). C1 and C2 are sites in Glen Clunie, S1 and S2 are sites in Glen Shee.

changes in calluna and grass cover

Protection from grazing for 6 years had a major impact on Calluna cover. There were significant effects of both site (F_3,8 = 4.63, P < 0.05) and fencing (F_1,8 = 348.54, P < 0.0001) on Calluna cover. In the fenced plots, cover increased on all sites by up to 20% by 1999, with Calluna at sites C1 and C2 (in Glen Clunie) increasing more rapidly than at site S2 (in Glen Shee). In contrast, on the plots exposed to grazing Calluna cover declined by 20–30% on all sites (Fig. 2a). There were no significant site–fencing interactions on Calluna cover. To a certain extent, changes in grass cover mirrored those in Calluna cover. Fencing had a significant effect on grass cover (F_1,8 = 172.47, P < 0.0001). There was also a significant site effect by 1999 (F_3,8 = 6.70, P < 0.05), but there were no significant site–fencing interactions. As with Calluna, changes in cover varied between sites, with the fenced plots at site S2 showing the smallest changes. Grasses performed particularly well on grazed plots, showing an average of 20–30% increase in cover at all sites after 6 years of the treatments (Fig. 2a).

Figure 2. The change in Calluna and grass cover between 1993 and 1999 at four moorland sites in north-east Scotland in response to fencing and nitrogen addition. C1 and C2 are sites in Glen Clunie, S1 and S2 are sites in Glen Shee. Mean values and standard errors are shown.

The fertiliser treatments showed rather more temporal and spatial variation than the effects of fencing. The effect of nitrogen addition on Calluna cover was significant (F_1,48 = 14.89, P < 0.001), but effects were initially small and site-dependant. Initially, Calluna cover increased on nitrogen-treated plots at sites C1 and C2 and declined slightly at sites S1 and S2 (data not shown), but by 1999 there was a decline in Calluna cover on nitrogen-treated plots at all sites (Fig. 2b). Phosphorus addition also had a negative effect on Calluna cover (F_1,48 = 13.26, P < 0.001), but only on plots that also received nitrogen (significant nitrogen–phosphorus interaction 1999: F_1,48 = 6.81, P < 0.05) (Fig. 3a). On average, Calluna cover decreased by 26% on plots receiving both nitrogen and phosphorus, but only by 1% on plots treated with phosphorus but not nitrogen. The response of grass cover to nitrogen addition was also initially rather small, but by 1999 grasses had increased on the fertilized plots at all sites (F_1,48 = 28.06, P < 0.0001), particularly at sites S1 and S2 where grass cover increased by over 20% (Fig. 2b). Grasses responded positively to the addition of phosphorus as well as nitrogen (F_1,48 = 14.78, P < 0.001), but the response was much greater on the plots that also received nitrogen (nitrogen–phosphorus interaction: F_1,48 = 6.62, P < 0.05) (Fig. 3b). There were no significant effects of potassium, either alone or in interaction with any other treatment, on the cover of either Calluna or grasses, even after 6 years of application.

Figure 3. The interacting effects of fencing, nitrogen and phosphorus addition on (a) Calluna and (b) grass cover between 1993 and 1999 on moorlands in north-east Scotland. C, control; N, nitrogen; P, phosphorus; and NP, nitrogen and phosphorus. Mean values and standard errors are shown.

Nitrogen addition decreased Calluna cover on grazed plots but increased it on plots protected from grazing, leading to a marginally significant fencing–nitrogen interaction (F_1,48 = 4.09, P = 0.05) at the end of the 6-year period. During this time, Calluna cover had declined by nearly 20% in the grazed without nitrogen plots and by over 40% in grazed plots with nitrogen added, such that Calluna was only increasing in cover on unfertilized plots that were protected from grazing. Overall, Calluna performed best on unfertilized plots that were protected from grazing, whilst grasses performed best on grazed plots where nutrients had been added.

calluna shoots grazed and calluna canopy height

Utilization rates on Calluna were high but variable. Over the 6 years of the treatments, there was a significant effect of both site (F_3,8 = 10.56, P < 0.001) and nitrogen addition on the percentage of shoots grazed (F_1,48 = 6.22, P < 0.05), but no other fertiliser treatment had any effect on grazing levels, nor were there any significant interactions between treatments. The mean percentage of current year's shoots that were grazed was 38% on plots without nitrogen addition, but this increased to 46% on plots receiving nitrogen. On average over the experimental period, between 35% and 39% of shoots were grazed at sites C1 and C2 (in Glen Clunie) and site S2 in Glen Shee, but 55% of shoots were grazed at site S1 (in Glen Shee).

Calluna height increased in the fenced plots between 1993 and 1999. Initially, the Calluna in the fenced areas was taller on the plots receiving nitrogen than on those that had not received nitrogen, but by 1999 the ungrazed Calluna was performing better in the unfertilized plots. Outside the fences Calluna height declined during the 6 years, particularly on the plots receiving nitrogen addition, such that by 1999 the Calluna in these plots was less tall than that in unfertilized plots. Over the 6-year period, mean Calluna height increased by 16 cm on fenced plots without nitrogen and 12 cm on fenced plots with nitrogen added. Outside the fences, Calluna did not increase in height at all on plots without nitrogen and declined by 2 cm on plots with nitrogen added. Site and fencing had a significant effect on the change in Calluna canopy height between 1993 and 1996 (site, F_1,3 = 4.42, P < 0.05; fencing, F_1,8 = 105, P < 0.0001). Calluna increased in height more in fenced plots, particularly at site S1. By 1999, nitrogen and phosphorus as well as fencing had a significant effect on the change in Calluna height since 1993 (fencing, F_1,8 = 181, P < 0.0001; nitrogen, F_1,48 = 15.71, P < 0.001; phosphorus, F_1,48 = 5.32, P < 0.05), but site had no effect. There was also a significant nitrogen × phosphorus (F_1,48 = 4.23, P < 0.05) interaction on the change in Calluna height between 1993 and 1999. Calluna height declined more on grazed plots with both nutrients added compared with plots where only nitrogen had been added, reflecting the fact that plots with both nitrogen and phosphorous added showed the largest decrease in Calluna cover and the largest increase in grass cover (Fig. 3).

effects on community composition

The effects of 6 years of fencing and fertiliser treatment on species composition were examined using the ordination technique RDA. Differences between the sites explained 24% of the variation in the plant community. Once the effect of site had been accounted for, the treatments explained 22% of the remaining variation (Table 1). The analysis was significant when tested using a Monte Carlo permutation test (P < 0.01). This suggests that by 1999 the species composition of the communities is beginning to diverge according to the grazing and fertiliser treatments they had been receiving over the previous 6 years.

Table 1.  Eigenvalues (a measure of the importance of the ordination axis) and intraset correlations (correlation between the ordination axes and the treatments (fertiliser and fencing)) from the RDA analysis

	Axis 1	Axis 2	Axis 3	Axis 4
Eigenvalues	0.123	0.069	0.016	0.006
Species-environment correlations	0.887	0.888	0.630	0.640
Percentage variance of species data	16.3	9.2	2.1	0.8
Percentage of species-environment relation	57.4	32.3	7.4	2.9
Intraset correlations
Fence	−0.901	−0.382	0.201	0.045
Nitrogen	0.338	−0.863	−0.195	0.322
Phosphorus	0.272	−0.1769	0.925	−0.120
Potassium	0.016	−0.281	−0.258	−0.924

The RDA ordination shows that Axis 1 is correlated with fencing (Table 1); fenced plots (small symbols) occur at the negative end and grazed plots (large symbols) at the positive end of Axis 1 (Fig. 4). Axis 2 is negatively correlated with fertiliser addition (Table 1), especially nitrogen addition, with the control plots (+ symbols) at the positive end of this axis, plots receiving potassium and/or phosphorus, but no nitrogen, in the middle of this axis (open symbols) and plots receiving nitrogen (and in some cases potassium or phosphorus as well) at the negative end of this axis (black symbols).

Figure 4. RDA ordination diagram of plots and key species in 1999. The main correlations between treatments and the axes are shown in bold at the end of each axis. F = Fence, N = nitrogen, P = phosphorus, K = potassium, C = control.

Dwarf shrubs, in particular Calluna vulgaris, Empetrum nigrum and Erica tetralix, were more common on ungrazed plots (Fig. 5). Vaccinium myrtillus was more common in plots with no fertilizer and no grazing. Bryophytes, such as Sphagnum sp., Hypnum sp. and Polytricum commune were generally more common on the plots without N (Fig. 5), with Hylocomium splendens occurring at the far positive end of Axis 2, indicating that this bryophyte in particular is adversely affected by N addition. The moss Rhytidiadelaphus squarrosus was more commonly found on grazed plots, occurring at the positive end of Axis 1. The wet moorland species Juncus squarrosus and Trichophorum cespitosum were more common on grazed plots, whereas Eriophorum angustifolium was more common on fenced plots. Sedges also differed in their response to the treatments, with Carex echinata and Carex nigra occurring more commonly in fenced plots and Carex binervis and Carex panicea occurring more commonly in grazed plots. Grasses, in particular Agrostis capillaris, Festuca ovina and Nardus stricta, and forbs, especially Galium saxatile and Potentila erecta, were more common on grazed plots than ungrazed plots (Fig. 5). The addition of nitrogen increased grass cover, but which species of grass depended on whether the plot was grazed: Festuca rubra and Deschampsia flexuosa increased in plots with nitrogen and protection from grazing and Festuca ovina, Agrostis capillaris, Agrostis canina and Nardus stricta increased with nitrogen and grazing (Figs 4 and 5).

Figure 5. RDA ordination of species in 1999. For simplicity only the position of the arrowheads for each species are shown. Species increase in cover in the direction of a line drawn from the origin to the arrowhead. 1, Agrostis capillaris; 2, Agrostis canina; 3, Anthoxanthum odoratum; 4, bare ground; 5, bare rock; 6, Calluna vulgaris; 7, Campanula rotundifolia; 8, Carex binervis; 9, Carex echinata; 10, Carex nigra; 11, Carex panicea; 12, Cerastium fontanum; 13, Deschampsia cespitosa; 14, Deschampsia flexuosa; 15, Dicranum scoparium; 16, Empetrum nigrum; 17, Erica tetralix; 18, Eriophorum angustifolium; 19, Eurhynchium praelogum; 20, Festuca ovina; 21, Festuca rubra; 22, Festuca vivipara; 23, Galium saxatile; 24, Hylocomium splendens; 25, Hypnum sp.; 26, Juncus squarrosus; 27, Lichen sp.; 28, Liverwort sp.; 29, Luzula multiflora; 30, Nardus stricta; 31, Plagiothecium undulatum; 32, Pleurozium schreberi; 33, Polygala serpyllifolia; 34, Polygala vulgaris; 35, Polytrichum commune; 36, Potentila erecta; 37, Scleropodium purum; 38, Rhytidiadelphus squarrosus; 39, Salix repens; 40, Sphagnum sp.; 41, Trichophorum cespitosum; 42, Trientalis europea; 43, Vaccinium myrtillus; 44, Vaccinium vitis-idaea; 45, Viola rivinana.

RDA was used to test which of the fencing and fertiliser treatments significantly affected the species composition of the community (Table 2). Fencing (P < 0.05) and site (P < 0.01) both had a significant impact on community composition. There was also a significant interaction between site and fencing (P < 0.01). The effects of nitrogen and phosphorus on the community were also both significant (P < 0.001), but the effect of potassium was not. The interactions between the fertiliser treatments and fencing were also tested, but none were significant. However, there was a significant interaction between site and nitrogen, suggesting that the impact of nitrogen on the species composition of the community varied between sites.

Table 2.  Significance of treatments and their interactions as tested by 15 partial RDA. P-values are Monte Carlo significance levels (9999 permutations). *P < 0.05, **P < 0.01, ***P < 0.001, NS = not significant

	Source	P
Block	Site	***
Main treatment	Fencing	*
Block–main treatment interactions	Fencing × site	**
Sub-treatments	N	***
	P	**
	K	NS
	N–P interaction	NS
	N–K interaction	NS
	P–K interaction	NS
Sub-treatments × main treatment interactions	Fencing × N	NS
	Fencing × P	NS
	Fencing × K	NS
	Site × N	**
	Site × P	NS
	Site × K	NS

analysis of plot scores for vascular plants for ellenberg, suited species and grime csr values

The effect of site

Site had a significant effect on Ellenberg fertility (F_3,8 = 12.91, P < 0.01), moisture (F_3,8 = 9.36, P < 0.01) and suited species grazing scores (F_3,8 = 17.18, P < 0.001) (Table 3). Contrast tests showed that plots at S1 and S2 had significantly higher Ellenberg fertility scores and lower moisture scores than those at C1 and C2 (P < 0.0001 for all contrasts). S1 and S2 plots had significantly higher grazing scores than C1 and C2 plots (P < 0.05 for all contrasts). Thus Glen Clunie sites were characterized by species unable to tolerate high grazing pressure and some Glen Shee plots had a greater cover of species associated with high soil fertility than those at Glen Clunie. Site also had a significant effect on Grime's CSR competitiveness score (F_3,8 = 7.57, P < 0.05), with C1 and C2 having a greater cover of competitive species than S1 and S2.

Table 3.  The effect of fertiliser addition and fencing on Ellenberg, suited species and Grime CSR scores. Only significant results are shown. *P < 0.05**P < 0.01***P < 0.001. F = Fence, P = phosphorus added, N = nitrogen added

	Site	F	N	P	N × P	N × K	F × N	F × P	Site × N	Site × F
Ellenberg
Moisture	**		*				*
Fertility	**	***	***	***	**			*
Suited species
Grazing	***	***	**				**		***
Grime CSR
C	*	*	*			*		*	**	*
S			*	***	*			***
R			***	***	***			*

The effect of fencing

Fencing had a significant effect on suited species grazing score (F_1,8 = 28.52, P < 0.001), Ellenberg fertility scores (F_1,8 = 30.02, P < 0.0001) and CSR scores for competition (F_1,8 = 6.77, P < 0.05) (Table 3). Unfenced plots had a greater cover of plants able to tolerate grazing and associated with high nutrient availability than fenced plots (P < 0.001 for all contrasts), while fenced plots had a greater cover of plants that were competitors than unfenced plots (P < 0.0001 for both contrasts).

The effect of nutrient addition

Nitrogen addition had a significant effect on Ellenberg fertility (F_1,48 = 42.56, P < 0.0001) and moisture scores (F_1,48 = 5.47, P < 0.05) and on suited species grazing scores (F_1,48 = 11.28, P < 0.01) (Table 3). Plots with nitrogen added had greater cover of plants associated with higher soil fertility (P < 0.0001) and able to tolerate higher grazing pressure (P < 0.0001), and a lower cover of plants able to grow in moist conditions (P < 0.05) than plots without nitrogen. Nitrogen addition also had a significant effect on all three CSR scores: competitive (F_1,48 = 6.72, P < 0.05), stress tolerant (F_1,48 = 7.19, P < 0.01) and ruderal (F_1,48 = 21.28, P < 0.0001). Nitrogen-treated plots had less cover of competitive and stress-tolerant species (P < 0.05 for both contrasts), but greater cover of ruderal species (P < 0.0001). Phosphorus addition had a significant effect on plot scores for Ellenberg fertility (F_1,48 = 43.16, P < 0.0001) and on CSR scores for stress tolerators (F_1,48 = 23.94, P < 0.0001) and ruderals (F_1,48 = 31.81, P = 0.0001). There was a greater cover of plants requiring high fertility levels (P < 0.01) that are classified as either ruderals (P < 0.0001) or stress-tolerant species (P < 0.0001) on phosphorus-treated plots than on plots without phosphorus. Potassium had no significant effect on the plot scores for any of the Ellenberg, suited species or CSR scores.

There was a significant nitrogen–phosphorus interaction for stress-tolerant (F_1,48 = 6.40, P < 0.05), ruderal (F_1,48 = 13.56, P < 0.001) and Ellenberg fertility scores (F_1,48 = 9.44, P < 0.01) (Table 3). Plots receiving both nitrogen and phosphorus had a lower cover of stress-tolerant species and a higher cover of ruderals and plants associated with high nitrogen levels than either plots with nitrogen (P < 0.001) or phosphorus (P < 0.001) individually.

interactions between grazing and nutrient addition

There was a significant interaction between fencing and nitrogen addition on suited species grazing scores (F_1,48 = 9.75, P < 0.01) and Ellenberg moisture scores (F_1,48 = 6.95, P < 0.05) (Table 3). Unfenced nitrogen-treated plots had significantly greater cover of plants able to tolerate high grazing pressure than the other combinations of fencing and nitrogen (P < 0.001 for all contrasts). Fenced plots with no nitrogen addition had greater cover of moisture-loving plants than other combinations of nitrogen × fencing treatments. There was a significant interaction between fencing and phosphorus treatments on Ellenberg fertility (F_1,48 = 4.60, P < 0.05) and on all three CSR scores (competitive, F_1,48 = 6.98, P < 0.05; stress tolerant, F_1,48 = 15.29, P < 0.001; ruderal, F_1,48 = 6.98, P < 0.05). Unfenced plots with phosphorus had a greater cover of plants with high Ellenberg fertility scores and of those that are classified as ruderals, and a lower cover of plants classified as stress tolerators than other combinations of fencing and phosphorus addition (P < 0.0001 for all contrasts). Unfenced plots with no phosphorus addition had a lower cover of competitive species than either fenced plots with or without phosphorus addition (P < 0.05 for all contrasts).

interactions between site and fertilizer or fencing

There was a significant interaction between site and nitrogen addition for suited species grazing scores (F_1,8 = 10.74, P < 0.0001) and CSR competitiveness scores (F_1,8 = 4.86, P < 0.01) (Table 3). Plots at S1 and S2 with nitrogen had a greater cover of grazing-tolerant species than plots with nitrogen addition at C1 or C2. Plots at C2 without nitrogen addition had a greater cover of competitive species than any plots at S1 or S2 with or without nitrogen addition. There was a significant interaction between fencing and site for CSR competitive scores (F_1,8 = 4.36, P < 0.05), with S1 unfenced plots having a lower cover of competitive plant species than any fenced or unfenced plots at C1 and C2 and any fenced plots at S1 or S2 with or without nitrogen addition.

Discussion

removal of grazing

Excluding grazing significantly benefited Calluna, whilst Calluna cover declined on grazed plots. The grazing levels at all four sites were sufficient to reduce cover, even on unfertilized plots. Grazing opens up the Calluna canopy, allowing grasses that would otherwise be over-topped and out-competed to successfully invade (Hartley 1997; Alonso et al. 2001). The species best able to do this are those either resistant to grazing because of their low palatability, such as Nardus stricta (Welch 1986), or those that tolerate high grazing pressure due to rapid regrowth capability, such as Agrostis capillaris. These species are the ones best correlated with the changes in community composition we observed in response to grazing (Fig. 4). Our results support the theory that herbivory, or removal of it, can relax or even reverse the competitive relationships between plant species in natural communities (Crawley 1997). Fenced plots had greater cover of competitive species that were grazing intolerant and able to grow in sites with low soil fertility than unfenced plots. Differential tolerance to grazing of co-occurring plant species, conferred by physiological and morphological traits, appears to be an important determinant of the responses of plant communities to herbivory (Augustine & MacNaughton 1998).

High-altitude heather moorlands are considered to be a slow growing plant community thought to be relatively resistant to change (Milne & Hartley 2001), but we recorded large and rapid changes; within 6 years up to 72% of the Calluna was lost on one grazed and fertilized plot, with most grazed and fertilized plots losing between 40 and 50%. The nutrient addition treatments were high (and only applied as two doses annually) and the sites were heavily grazed, which may explain the dramatic responses. Clearly complete suspension of high grazing pressures produces impressively rapid changes in Calluna cover, indicating that the increase in graminoid cover brought about by over-grazing can be reversed once grazing is removed (Hartley 1997; Hartley et al. 2003). However, this may not help us predict the responses of the community to grazing reduction, rather than absence. In this experiment, both sheep and deer grazing were removed; if only sheep grazing were removed, the change in vegetation composition may not be as dramatic, as increased grazing by red deer may offset the decline in grazing by sheep (Hope et al. 1996). At present it seems easier to predict the effects of increases in the numbers of sheep and red deer on plant communities than to predict the effects of reductions in herbivore loads (Grant & Maxwell 1988; Milne & Hartley 2001). An additional complication is that there was large spatial and temporal variability in the responses of Calluna cover to the removal of grazing, even within this single vegetation type at adjacent sites. Rates of change in Calluna cover in response to grazing clearly depend on the intensity of the previous grazing pressure and site-based factors such as soil type (see below).

effects of fertiliser

Fertiliser application produced similarly dramatic changes in the community composition as grazing, in part because fertiliser addition increased the shoot grazing rate on Calluna (Duncan et al. 1994). The higher grazing rate on fertilized plots may not solely be due to fertilized Calluna being more attractive; there is also evidence that grazers are attracted to small patches of grass, particularly if it is nitrogen rich, and then graze the neighbouring Calluna (Palmer et al. 2003). One notable result is that in the absence of grazing, nitrogen does not lead to a significant loss of Calluna cover, even after 6 years of fertiliser addition to the fenced plots (Fig. 2). Thus in the absence of grazing Calluna can tolerate or even benefit from increased levels of nitrogen, at least in the short term.

Nitrogen had a greater effect than phosphorus or potassium on overall community composition and resulted in an increase in cover of grazing-tolerant species associated with fertile and dry sites. Agrostis sp. seemed to be the main beneficiary of increasing nitrogen addition. Similar results have been found in the Netherlands, where nitrogen addition also shifts the competitive balance in favour of grasses (Heil & Diemont 1983; Aerts & Berendse 1988), although some workers here have also found that fertilization does not lead to an immediate detrimental effect (Prins et al. 1991). The competitive ability of plants is known to change depending on the availability of resources (Crawley 1997). At low nutrient levels, Calluna vulgaris can out-compete grasses such as Nardus stricta and Molinia caerulea (Heil & Bruggink 1987; Aerts et al. 1990; Hartley & Amos 1999), but when fertiliser is added to Calluna vulgaris and Nardus stricta grown in competition, Nardus stricta out-competes Calluna vulgaris (Hartley & Amos 1999). However, other studies have found that under some conditions Calluna out-competes grasses even at high nutrient levels (Aerts et al. 1990; Genney et al. 2000, 2002).

The effects of fertiliser addition are temporally variable and as one nutrient is added it may cause a second to become limiting over time (Carroll et al. 1999). Hence, atmospheric nitrogen deposition on nitrogen-limited heather moorland has been shown to move the community from nitrogen limitation to being phosphorus limited (Berendse & Aerts 1984; Kirkham 2001). Our results confirm that some effects of nitrogen addition may alter over time as phosphorus becomes limiting instead; this interpretation is supported by the fact that there is a significant nitrogen–phosphorus interaction on Calluna cover in 1999, but not in 1996 after only 3 years of treatment. Further, phosphorus addition only had a negative effect on Calluna cover and a positive effect on grass cover on plots that had also received nitrogen, suggesting that the benefits of phosphorus for grasses relate to nutrient limitation arising from increased growth in response to the nitrogen addition. Other studies have also found that the addition of phosphorus on its own does not lead to the replacement of ericoid species by grass species (Heil & Diemont 1983), but it can alter the competitive balance in favour of Molina when in competition with Erica tetralix (Berendse & Aerts 1984; Aerts & Berendse 1988). Potassium had very few significant effects either alone or in combination, suggesting this nutrient rarely becomes limiting for this plant community.

differences between glens and the effect of different soil types

The effects of grazing and fertilizer varied markedly between the four sites. The RDA analysis showed a significant site–nitrogen interaction affecting community composition. Calluna recovered more rapidly in fenced plots in Glen Clunie (sites C1 and C2) than in Glen Shee (S1 and S2), possibly because the higher grazing rates in Glen Shee meant the Calluna was in poorer condition when the fences were erected. However, soil type may also play an important role in explaining why the relative competitive balance between Calluna and grasses differs between glens. For example, the greater adverse impact of nitrogen addition on Calluna cover in Glen Shee may be due to the fact that the drier, more mineral soils at sites S1 and S2 are already less favourable for Calluna growth then the soils at sites C1 and C2, which are wetter and contain more organic matter (Fig. 1), and nitrogen addition is thus even more effective at altering the competitive balance in favour of grasses instead of ericoid species (Berendse & Aerts 1984). Nitrogen addition seems to produce larger changes in nitrate and ammonium availability at site S2 than at site C1 (Fig. 1), although there was no overall significant interaction between site and mineralization rates.

In upland ecosystems, inorganic inputs from wet and dry nitrogen deposition are likely to be significant in relation to the quantities of nitrogen cycling through the plant-microbial-soil system. As a result, atmospheric nitrogen deposition can alter the balance between soil microbial activity, the availability of nutrients such as phosphorus, soil acidity and carbon turnover (Williams & Anderson 1999). The underlying soil type will influence the way the nitrogen is cycled within the ecosystem, rates of nitrification and denitrification and any changes in litter decomposition and microbial activity (Yesmin et al. 1996; Williams & Anderson 1999), all of which drive changes in the availability of nutrients.

changes in community composition

The most abundant grass species increasing at our sites as Calluna dominance was lost on fertilized plots were Agrostis sp., Nardus stricta, Deschampsia flexuosa, Festuca ovina and Festuca rubra. Which of these grass species increased in response to fertiliser depended on whether grazing was present or not: Festuca rubra and Deschampsia flexuosa increased with nitrogen addition on fenced plots, and Festuca ovina, Agrostis capillaris and Nardus stricta increased with nitrogen addition on unfenced ones. This reflects the palatability of the species to grazers: D. flexuosa is preferred to F. ovina and N. stricta (Welch 1986; Alonso et al. 2001) so it is best able to replace Calluna in the absence of grazing, whereas the less palatable and more tolerant grasses perform relatively better in the presence of grazing. Similar results have been reported from the Netherlands, where the addition of nitrogen leads to the replacement of Calluna with grass species, usually Molina caerulea (Aerts et al. 1990), but also Festuca ovina (Heil & Diemont 1983) and Deschampsia flexuosa (Diemont 1996). Although we found that grass cover increased more rapidly at the Glen Shee sites, reflecting the more favourable soil type for grass growth, there was no difference in the number of grass species present between the two glens in 1999.

The results presented here highlight two important issues for the prediction of the responses of upland plant communities to nitrogen deposition. First, although nitrogen addition clearly has a major effect on moorland vegetation composition, the presence of grazing alters both the direction and the rate of this change, which needs to be taken into account when calculating critical loads for this community. In this study the nitrogen addition was considerably more than the estimated critical load for heather moorland of 10–20 kg N ha⁻¹ year⁻¹ (Acherman & Bobbink 2003), but Calluna did not decline significantly in the absence of grazing even after 6 years of treatment. Several other studies have found no decline in Calluna cover in response to nitrogen addition at low grazing levels within this time frame (Prins et al. 1991; Power et al. 1998; Carroll et al. 1999). Secondly, the impact of nitrogen deposition on the community composition varied significantly with site. If the plant communities on two neighbouring moors show such different responses to nitrogen deposition, this may have implications for the general applicability of critical loads. For example, these results show that the addition of the same amount of nitrogen has a more adverse effect on Calluna cover at sites with mineral soil and heavy grazing than at sites with peaty soil and lighter grazing, suggesting that both soil type and grazing intensity may influence critical loads. Studies on other upland plant communities have also found interactions between the impacts of grazing and nutrient addition, so this may be an issue of general concern when considering the conservation of these habitats (Hartley et al. 2003; Van der Wal et al. 2003).

grazing impacts and resource availability

Some of the greatest changes in community composition were on those plots receiving both grazing and nitrogen addition. This was especially noticeable at sites C1 and C2, where the unfenced nitrogen plots were dominated by grassland species, whilst the fenced nitrogen plots were dominated by species typical of wet moorland and dwarf shrub heath. These changes in species composition were due to the interaction between nutrient availability and grazing pressure, driving changes in the competitive ability of the plants. This change in competitive ability concurs with other work studying the effect of grazing along a nutrient gradient (e.g. Turkington et al. 1993; Van der Wal et al. 2000).

The basis for species interaction in our moorland system may be the same as in the models of Husiman et al. (1999), namely asymmetric competition for light. Grazing decreases competition for light (Alonso & Hartley 1998; Alonso et al. 2001), as has been observed in other systems where plant stature has been shown to be the key to explaining herbivore impacts on vegetation change (e.g. Van der Wal et al. 2000). The importance of light in the competition between Calluna and grasses has been suggested as being more important than the competition for nutrients (Aerts et al. 1990). However, nutrient availability clearly plays a role, as shown by the effects of nutrient addition on the community composition. Models predict that palatability and nutrient availability interact, such that edible plants are better competitors in the absence of herbivores but worse in their presence (Grover 1995). Again our results support this prediction. For example, we observed palatable grasses (e.g. Deschampsia flexuosa, Festuca rubra and Agrostis stolonifera) increasing on fertilized ungrazed plots and unpalatable species (Nardus stricta and Festuca ovina) increasing on fertilized grazed plots. Tolerance to grazing also plays an important role though, as very palatable species can still compete successfully for added nutrients in the presence of grazing if they have high regrowth potential, hence the increase in Agrostis capillaris on fertilized grazed plots.

It is often hard to disentangle the influence of competition alone from the direct impacts of nutrients and disturbance (Turkington et al. 1993). According to the CSR triangle model developed by Grime (1977, 1979), the importance of competition in determining community composition decreases as the importance of disturbance (e.g. grazing) increases. In undisturbed areas the percentage cover of a species is in part determined by its neighbours (competition), whereas in highly disturbed areas the cover of a species is due to the direct impact of the disturbance. The sites in this study are clearly between these two extremes, with both competition (for nutrients) and disturbance (grazing) driving changes in the community composition by altering the outcome of competition between grasses and ericoid species (Berendse 1985; Alonso et al. 2001). Defoliation prevents Calluna vulgaris from increasing its above-ground biomass in response to fertiliser, but this is not the case for Nardus stricta, which can increase in above-ground biomass in response to fertiliser largely irrespective of whether it is defoliated (Hartley & Amos 1999). This may also be true for other graminoid species, suggesting a combination of grazing and nutrient inputs may provide the most favourable conditions for grasses to outcompete Calluna vulgaris. This may explain why the greatest changes occur on these plots. Thus in the short term, a reduction in grazing is likely to be more beneficial in promoting the recovery of overgrazed moorland than a reduction in nitrogen deposition, but in the long term even ungrazed Calluna vulgaris may decline due to increased nutrient addition. This work therefore agrees with previous studies suggesting that critical loads for nitrogen deposition may need to take account of the interacting effects of grazing and nitrogen deposition (Hartley et al. 2003; Van der Wal et al. 2003). The importance of grazing, and other forms of management, is now clearly recognized in the latest assessments of critical loads (Acherman & Bobbink 2003).

ecological specialization of plants in response to changes in nutrient availability and grazing

Our experiment supported predictions based on known ecological strategies, in that plants intolerant of grazing and disturbance and associated with low nutrient demands occurred on the fenced plots, whilst plants tolerant of grazing and disturbance and associated with higher nutrient levels occurred on the unfenced plots. Grime (1977) defined competitive plants as those growing in situations with low stress and low disturbance. Unfenced plots are associated with a higher level of disturbance due to grazing and trampling than fenced plots, thus one would expect to find plants associated with low levels of disturbance (competitors) in the fenced plots. More nutrients will be available to plants on the grazed plots, due to animal dung, than on fenced plots, therefore plants with higher nutrient demands are also likely to be found on the unfenced plots.

Plots with nitrogen addition had a greater cover of plants associated with higher levels of nutrients than untreated plots and a lower cover of stress-tolerant and competitive species. The addition of nutrients reduced one of the stresses of growing in the moorland environment, that of low nutrient supply, and therefore plants that were not adapted to cope with this stress could survive in these plots. Competitive species, those adapted to utilizing all available nutrients, water and space, also declined in fertilized plots as this adaptation no longer gave them an advantage. Thus the addition of fertilizer no longer required the plants to have this competitive or stress-tolerant characteristic in order to survive. Plots with nitrogen and/or phosphorus additions were characterized by having a greater cover of ruderal species than plots without these treatments. Ruderal species are able to establish quickly and grow rapidly, utilizing the additional nutrients.

The changes in community composition caused by changes in grazing pressure and the addition of fertilizer can thus be explained in terms of the ecological tolerances of individual species. This allows us to indicate the types of plants that are likely to increase or decrease in cover following changes in grazing regimes or nutrient availability. However, we also found that site was an important influence and could modify our predictions: whilst plants with a certain type of ecological specialization may consistently increase or decline in cover in response to grazing and/or nutrient addition, the actual species may vary according to the site at which these treatments are imposed.

Conclusion

Protection from grazing caused a rapid and dramatic increase in Calluna cover whereas grazing caused a decline in Calluna cover, particularly when combined with nitrogen addition. However, in the absence of grazing, substantial nutrient addition did not lead to a significant decline in Calluna cover. Grazed plots had a greater cover of grass and forb species, while ungrazed plots had a greater cover of dwarf shrub species. Bryophytes were generally more common on plots without nitrogen addition, than those with nitrogen addition. Nitrogen addition resulted in an increase in grass cover, although which species increased depended on whether the plot was grazed. Fencing increased the cover of grazing-intolerant, low nutrient demanding, competitive plants. The nitrogen and phosphorus plots had significantly higher cover of ruderal species that required high fertility and were able to tolerate high grazing pressure. Thus changes in plant community composition following changes in grazing pressure and increases in nutrient addition can be explained by the competitiveness and grazing tolerance of the species. The interacting effects of grazing and nutrient addition on this community related well to the predictions of recent models (Grover 1995; Huisman et al. 1999), suggesting the important processes in structuring moorland plant communities are asymmetric competition for light, the impacts of generalist herbivores on tall dominants, and the relative competitive abilities of palatable and unpalatable grass species at different grazing intensities.

Acknowledgements

We thank the James Weir Foundation and The Royal Society Edinburgh for funding (S. E. H). The Owner, Factor and Gamekeepers at Invercauld Estate kindly allowed us to use their land and erected the fencing. Many people have helped with the fieldwork for this project, especially Anne Davies, Caroline Young, Gudrun Heckemeier, Ian Paterson, Isabel Alonso, Jemma Gornall, Joanne Harvey, Keith Marshall, Kenny Nelson, Louise Amos, Maddie Thurlow, Nigel Watson, Rory Gordon, Stuart Munro and Susie McIntosh. We thank them all. Bridgett Emmett at CEH Bangor carried out the soil analysis. We are also grateful to David Elston of BIOSS for statistical advice. Steve Palmer and Rene van der Wal kindly commented on earlier versions of this manuscript.