Quantifying the grazing impacts associated with different herbivores on rangelands


S.D. Albon, Macaulay Institute, Craigiebuckler, Aberdeen AB15 8QH, UK. E-mail s.albon@macaulay.ac.uk


  • 1Rangelands, produced by grazing herbivores, are important for a variety of agricultural, hunting, recreation and conservation objectives world-wide. Typically, there is little quantitative evidence regarding the magnitude of the grazing impact of different herbivores on rangeland habitats to inform their management.
  • 2We quantified the grazing and trampling impact of sheep, cattle, red deer Cervus elaphus, rabbits Oryctolagus cuniculus, mountain hares Lepus timidus and red grouse Lagopus lagopus on open-hill habitats in 11 areas of upland Scotland. The degradation of heather in upland Scotland Calluna vulgaris-dominated habitats, of conservation significance at a European scale, has been attributed, anecdotally, to increasing sheep and red deer populations.
  • 3Field indicators of habitat condition were used to generate a five-point scale of impact in vegetation polygons of seven habitats. The presence of each herbivore species was attributed on the basis of ‘signs’ of occupancy. A Bayesian regression model was used to analyse the association of herbivore species with grazing impact on plant communities, controlling for environmental attributes.
  • 4Overall the presence of sheep was associated with the largest increase (7/11 areas) in grazing and trampling impact of all herbivores. Cattle had the second largest impact but generally this was restricted to fewer areas and habitats than sheep. In contrast, impacts associated with wild herbivores tended to be small and only significant locally.
  • 5Although red deer presence was associated with a significantly lower impact than sheep, this impact increased with increasing deer density at both land-ownership and regional scales. For sheep there was little or no evidence of density dependence.
  • 6Synthesis and applications. The higher impact associated with sheep presence probably reflects their greater aggregation because of their limited ranging behaviour, exacerbated by sheep being herded in places convenient for land managers. Consequently, future reductions in sheep numbers as a result of reform of European Union farming policies may limit the extent of their impact, but not necessarily the local magnitude. However, reductions in sheep stocks may lead to increases in deer densities, with greater impact, particularly in heather-dominated habitats. Where habitat conservation is a priority this may well require a reduction in deer numbers.


Across many parts of the world, vertebrate herbivores influence the structure, composition and functioning of ecosystems (Hobbs 1996; Augustine & McNaughton 1998). The rangelands that result from grazing and browsing are an important resource, managed for a variety of agricultural, forestry, hunting, recreation and conservation objectives (Gordon, Hester & Festa-Bianchet 2004). However, while there has been much progress in collaborative management in recent times, these multiple objectives can lead to conflicts, including habitat degradation as a result of over-grazing (Hallanaro & Usher 2005; Mysterud 2006). For example, in upland Scotland since the Second World War the extent of heather-dominated vegetation communities and semi-natural grasslands has declined markedly (Tudor & Mackey 1995), primarily because of afforestation and agricultural reclamation (Miles 1988). A direct consequence of these changes in land use has been increasing concentrations of stocks of sheep and red deer on the remaining habitat, highlighting concerns about grazing and trampling impacts, in particular to dwarf-shrub heath, blanket bog and montane plant communities (Sydes & Miller 1988; Staines, Balharry & Welch 1995). While there has been a growing awareness of the national and international importance of such habitats for the conservation of biodiversity (Thompson et al. 1995), there has been no previous attempt to quantify simultaneously the relative grazing impacts associated with different herbivore species using these rangelands (Clutton-Brock, Coulson & Milner 2004).

The total sheep stock in Scotland increased from 6·9 million (M) in 1945 to 8·6 M in 1965, before declining by about 12% to 7·5 M over the next decade (annual agricultural statistics, Department of Agriculture for Scotland and Department of Agriculture and Fisheries for Scotland, UK). The size of the national flock rose again after the UK joined the European Common Market in 1973. to peak at more than 9·4 M in the early 1990s, although in some parts of the country numbers did not exceed the 1965 peak (annual agricultural statistics, Scottish Office Agriculture and Fisheries Department and Scottish Executive, UK). In upland Scotland most sheep are kept on enclosed, improved pastures at lower altitudes for much of the year. However, in summer the widespread practice is to allow sheep flocks free-range access to the semi-natural vegetation above the enclosed land. In some areas cattle may also be given access to the same unenclosed land, although they tend to roam less widely. Typically these open-hill rangelands are privately owned and often managed as sporting estates, with red deer being the main quarry species across the country (Wightman & Higgins 2001). In central and eastern Scotland, many estates are managed primarily for red grouse shooting. Here mountain hares may be sufficiently numerous to be hunted as well, but usually on a less systematic basis (Hewson 1976).

Estimates of red deer population sizes were not collected systematically in Scotland until 1960, following the establishment of the Red Deer Commission (now Deer Commission for Scotland). Analysis of repeat counts at landscape scales (deer subpopulations, 100–2000 km2) suggest that from the mid-1960s red deer increased steadily to average 14 deer km−2 in 1986, 40% higher than in 1961 (Clutton-Brock & Albon 1989). In 1986, deer densities differed between areas of Scotland by an order of magnitude, with both the spatial and temporal variation in density correlating negatively with sheep stocks (Clutton-Brock & Albon 1989). Red deer numbers may still be growing but both the rate of this increase and their impact on the natural heritage are disputed (Hunt 2003; Clutton-Brock, Coulson & Milner 2004).

In this study, we quantified the grazing and trampling impacts associated with six different herbivore species on seven semi-natural, open-hill habitats, in 11 upland areas of Scotland. The five-point scale impact data we analysed were recorded at the vegetation polygon (patch) scale (down to 250 m2) of the different habitats (Brewer et al. 2004). First, we described how the estimated grazing impact scores were associated with the presence/absence (not recorded present) of sheep, cattle, red deer, mountain hares, rabbits and red grouse recorded at the same patch scale as plant communities.

Secondly, we explored whether it is possible to detect differences in grazing impact associated with the presence of sheep vs. red deer. Given the two species have similar feeding niches (Milne et al. 1976) because of their similar body size and gut morphology, it would initially appear problematic to detect differences in grazing impact. However, Hofmann (1989) categorized sheep as ‘grazers’ and red deer as ‘intermediate feeders’ selecting a diet of browse and graze. Research investigating detailed foraging behaviour in heather–grass mosaics suggested that, compared with deer, sheep are more selective of grass species, but at grass–dwarf-shrub heath boundaries sheep have similar impacts on heather as deer, particularly where the patchwork is fine grained (Clarke, Welch & Gordon 1995; Hester et al. 1999). In general, however, red deer grazing impacts on heather are more diffuse (Palmer et al. 2003).

Thirdly, we investigated the relationships between estimated impact on vegetation and the stocking rates of sheep and counts of red deer for land management units (10–100 km2) available for two areas of Scotland. Finally, we demonstrated that estimates of mean impact within habitats at the largest landscape scales (> 500 km2) are related to the mean density of deer in these different areas across Scotland.

The implications of these results are discussed in terms of a more integrated approach to the management of herbivores on rangelands, recognizing the potential conflicts arising from contrasting objectives of different land managers. In addition, we discuss future research priorities to provide the evidence base for adaptive management targeted at more specific outcomes, in particular the enhancement of biodiversity.

Materials and methods

grazing and trampling impact assessment

Grazing and trampling impacts were assessed in 11 deer management group (DMG) areas between 1997 and 2003 (Fig. 1), following a standard method of surveying the impacts of grazing, browsing and trampling by larger herbivores in upland habitats (MacDonald et al. 1998). This approach provides a means of recording the state of habitats using descriptive classes, focusing on directly observable effects. The methodology uses a series of field indicators for each habitat type, including biomass removal, sward height and structure, selectivity of grazing, accumulation of plant litter, physical damage and dung.

Figure 1.

The name and location of the 11 deer management group (DMG) areas in which herbivore grazing and trampling impacts were surveyed (year of survey after the name) and the distribution of impact scores (L, light; L/M, light/moderate; M, moderate; M/H, moderate/heavy; H, heavy) for n= vegetation polygons sampled within each DMG. The percentage of vegetation polygons in each habitat is shown in Table 1.

For each field indicator, such as the proportion of long shoots of heather browsed and the extent of trampling damage, a number of alternative states were described, relating to light, moderate and heavy (L, M, H, respectively) impacts (see Table S1 in the supplementary material). The standard field survey sample area of upland habitat was 0·25 km2 (25 ha). Each indicator was assessed separately, based on a number of point estimates, and an overall assessment, derived for a particular habitat type, was averaged across all indicators. As all three impact classes (L, M and H) could be observed within a part-polygon, a method of summarizing the impact across a habitat was devised that took into account the spatial heterogeneity. This was based on the percentage of the area occupied by each impact class (see Appendix S1 in the supplementary material) and had the effect of smoothing the three-class impact scale into a more continuous five-point scale by introducing intermediate classes (L/M and M/H). Where a number of discrete plant communities occurred within the sample area, they were all assessed separately.

The impact assessments were made for seven upland open-hill habitats based on the Land Cover Scotland 1988 (LCS88) data set (MLURI 1993). (i) Blanket bog, dominated by common cotton-grass Eriophorum vaginatum L., bog mosses Sphagnum spp. and heather. (ii) Dwarf-shrub heath, either dry heath, dominated by heather or blaeberry Vaccinium myrtillus L., or wet heath, dominated by cross-leaved heath Erica tetralix L. and deer-grass Trichophorum cespitosum L. Hereafter, these are both called Heath. (iii) Coarse grassland, characterized by species such as mat-grass Nardus stricta L., purple moor-grass Molinia caerulea L. and tufted hair-grass Deschampsia cespitosa L. (iv) Montane Grassland, wind-clipped grass/sedge/moss heath, with a range of species including stiff sedge Carex bigelowii L., blaeberry, woolly fringe moss Racomitrium lanuginosum L., and a variety of smooth grasses. (v) Montane heath, wind-clipped heath, dominated by heather. (vi) Smooth grassland, dominated by common bent Agrostis capillaries L. and sheep's-fescue Festuca ovina L. (bent-fescue grassland). (vii) Smooth grassland with bracken or rushes, dominated by bent-fescue grassland with bracken Pteridium aquilinum L. and/or rush Juncus spp.

Three of the 11 DMG (Table 1) were sampled on a 0·25-km2 complete-coverage basis, where assessments were made for all the part-polygons of the main habitats present that occupied more than 10% of any 0·25-km2 sample area (Brewer et al. 2004). In the other eight DMG, 0·25-km2 sample squares were selected in a random stratified approach to provide data for management planning over extensive areas (200–1000 km2) in a more rapid and cost-effective manner (Nolan et al. 2003). A geographical information system (GIS)-based computer model was used to select 0·25-km2 sample areas of different habitats randomly, ensuring that the strata also took account of land-management units within and between estates. In practice this reduced field sampling to between 12% and 21% of the total area (Table 1).

Table 1.  Summary survey details for each deer management group (DMG). In 1997 and 1998, surveys were done in smaller DMG in which all 0·25 km2 were surveyed. From 1999 onwards, between 12% and 21% (median 17%) of the total area was surveyed. The percentage vegetation (BB, blanket bog; H, heath CG, coarse grassland; MG, montane grassland; MH, montane heath; SG, smooth grassland; SR, smooth grassland with bracken and/or rushes) is based on polygons and not on an area basis
DMGYearTotal area (km2)Sampling intensity (%)Number vegetation polygon% vegetation
West Sutherland20001136 14 83639·641·3 7·22·7 3·1 5·30·9
Northern19991318 12118245·538·6 4·80·0 1·9 4·05·2
North Ross2001 771 17 81935·342·3 5·04·2 5·6 5·71·9
East Sutherland20001006 17 97244·838·9 4·30·2 2·2 6·53·2
Gairloch1998 346100265133·753·8 0·72·2 7·2 1·90·5
Cairngorm–Speyside1997 420100204128·548·2 4·82·910·1 4·80·6
South Ross20001611 18207227·939·210·26·6 6·4 6·82·9
West Grampian2002 710 21 96926·541·210·54·5 6·8 8·62·0
Mid-West Association2003 629 16 71931·039·315·74·6 3·8 4·60·9
Angus Glens1999 620 21106714·942·813·93·8 2·512·19·9
South Loch Tay1998 148100343718·633·524·73·1 4·811·93·5

Aspect, altitude and slope were determined for each part-polygon in ArcGIS 8·3 using a digital elevation model, reproduced from map data by permission of Ordnance Survey (© Crown copyright reference MLURI GD27237X 2006), with a resolution of 50 × 50 m. An index of topographic exposure (TOPEX) was calculated for each part-polygon as the sum of angle to the skyline in degrees, for the eight cardinal directions. Dominant soil type for each part-polygon was determined from an overlay of soil types from the 1 : 250 000 survey of the soils of Scotland (Macauley Institute for Soil Research 1984). The presence of burning was recorded in the field for each part-polygon of vegetation, assessed on a four-point scale of no-burning and the presence of small (< 2 ha), medium (2–5 ha) or large (> 5 ha) burns.

herbivore presence

Within each part-polygon, the presence/absence (not recorded present) of each species of herbivore was attributed on the basis of a range of ‘signs’ of occupancy, including visual sighting, evidence of recent animal presence (strands of wool, hair or feathers on vegetation, lying-up areas, animal tracks, trampling and thrashing of heather and burrows, etc.) and the presence of dung. While it was recognized that the identification of some species based on dung (notably deer/sheep) can be problematic, this was very rarely used as the sole criterion for differentiating animal presence. Surveyors carrying out the impact assessments were experienced in the identification of animal dung and calibrated their observations, aided by field guides describing animal species from tracks, signs and dung. As there may have been some ‘false negatives’, where a species recorded absent had been present but went undetected, we refer to ‘recorded’ presence and absence.

sheep and deer density estimates

After 2000, information on livestock management (numbers, grazing regime, etc.) was collected at the estate (separate landowner/management units) level. Specifically for sheep, we sought data on numbers and the period they were free-ranging on the open-hill. Given that some flocks were out all year and others for different periods of the summer, we calculated densities as total year equivalents divided by the area of the estate and the proportion of vegetation polygons in which sheep were recorded present, and expressed this as number km−2. The most recent deer counts, between 0 and 3 years prior to the survey, were supplied by the Deer Commission for Scotland, divided by the unit land area, and expressed as number km−2. For comparisons with sheep, we divided the density by the proportion of vegetation polygons occupied. As the distribution of habitat types differed between DMG (Table 1), we investigated deer density as an explanatory variable for the geographical variation between DMG in estimated mean impact scores within habitats.

statistical analysis

The ordered categorical (ordinal, five-point scale) response variable was modelled as described in Agresti (1984), where impact is a continuous entity on a five-point scale. Mathematically, this translates to a partitioning of the real number line by four cut-points that separate out the five response classes: light, light/moderate, moderate, moderate/heavy and heavy. This is preferable to simply assigning numerical values to the impact classes and treating this as a continuous response variable, or fitting a multinomial model that ignores the ordering of the classes.

A Bayesian regression analysis accounting for the ordinal response, similar to that of Brewer et al. (2004), was used to identify explanatory variables. The response impact class from the ith part-polygon (e.g. i= 1, … , 2072 for Cairngorm–Speyside) in grid square j (j = 1, … , 1176 for Cairngorm–Speyside) could thus be assigned an integer value between 1 and 5, called Rij, that corresponded with the five classes listed above (in order). The impact class can be expressed in terms of a continuous latent variable, Yij (representing the underlying continuous grazing impact), taking values as follows:

Rij = k if
Yij ∈ [ak−1 (ak), k = 1, … , 5

where ak represents the cut-off points separating the classes, with a0 = −8 and a5 = 8. Further technical details of the statistical model can be found in the supplementary material. Unlike Brewer et al. (2004), variograms of the unstructured spatial effects suggested no need to account specifically for spatial autocorrelation.

The model included the following ecological and environmental variables as covariates. Herbivore factors: sheep, deer, cattle, hares, rabbits, grouse; with interactions sheep + deer, sheep + rabbits, deer + rabbits, hares + grouse (given the large number of potential interactions these were chosen as the most worthy of exploring). Habitat factors: vegetation community (seven levels). Interactions between herbivores and habitats. Other environmental factors, including: aspect (four levels, north, east, south and west); muirburn (four levels of burning); dominant soil categories (up to 21 levels). Environmental covariates: altitude; slope; TOPEX. Human-defined factors: estate/landownership unit.

Our model included this broad set of variables so that we could be more confident that estimated herbivore effects really were because of the herbivores and not unexplained ecological and environmental variables. OpenBUGS (Thomas 2004) was used to analyse the data fitting the above model. As there were a large number of explanatory variables, it was not always possible to fit all the terms, or all levels of all the factors, because certain combinations did not occur in the data sets for particular DMG. Thus some effects were not estimable; this was true of some of the interaction terms between herbivore presence and habitat where very small numbers of part-polygons recording presence (or absence) of the herbivore led to complicated patterns of aliasing. As a rule of thumb, the interaction term for a herbivore–habitat combination was not fitted if fewer than five part-polygons for that habitat were recorded as having the herbivore present (or absent). This was a particular problem for rabbits, where we were only able to estimate the grazing effects in five of 11 DMG. Also, in two of the DMG, North Ross and West Grampian, deer presence was recorded in nearly all part-polygons, and hence the effect of deer could not be estimated. Because the Bayesian approach was computer intensive, requiring many runs for different DMG and subsets of variables, we ran OpenBUGS on a Beowulf Linux cluster.

As the parameter estimates were on the logistic scale, and difficult to interpret, we found that a convenient way to represent the effects of the recorded presence compared with apparent absence of a herbivore species was the estimated change in the probability of observing an impact class of ‘Moderate’ or worse. For example, in the Northern (NO) DMG, the predicted impact of the presence of sheep on blanket bog was 0·42 (Fig. 3). This metric enabled a useful comparison in terms of comparing impact classes across DMG and habitat types, and was likely to be more useful than standard ‘proportional odds’ statements.

Figure 3.

The impact of the presence of deer (closed circles) and sheep (open circles) across the 11 DMG by habitat. The bars display 95% (2·5–97·5%) credible intervals. Where the mean is positive and the credible interval does not include zero, the probability of increasing impact with the presence of a particular herbivore is significant. If the credible interval touches zero then the significance is marginal. Where the credible interval includes zero, the overall impact is not significant because of the variance between polygons within habitats. The letters D and S beneath the bars denote that the herbivore by habitat interaction term was not estimable for that DMG and habitat for deer or sheep, respectively. The asterisks above the bars signify that there was a significant difference between the effects of sheep and deer for that DMG and habitat; these were all in the direction of sheep having a greater impact on vegetation than deer. It was not possible to estimate deer effects in North Ross (NR) or West Grampian (WG).


herbivore species distribution

Deer were the herbivore most frequently recorded as present in vegetation polygons in all DMG (median 90·1%, interquartile range 81·3–96·6%), except South Loch Tay (74·0%), where evidence of sheep occurred in 96·3% of vegetation polygons (Table 2). Overall sheep were the second most frequently recorded herbivore (median 42·5%, interquartile range 26·6–55·5%). Evidence of deer and sheep using the same vegetation polygons was common (median 31%, interquartile range 22·9–55·3%) but rarely were neither recorded (median 2·4%). Cattle were an order of magnitude less frequently recorded than sheep (median 2·9%), reflecting their comparatively low numbers.

Table 2.  The percentage of vegetation polygons (sample size given in Table 1) in each deer management group (DMG) area in which different herbivores were recorded. For each DMG the first four columns total 100%
DMG% with no deer or sheep% sheep only% deer only% both deer and sheep% cattle% hares% rabbits% grouse
West Sutherland5·7 3·967·622·71·8 1·1 0·5 8·1
Northern5·012·139·943·03·822·0 2·437·5
North Ross0·1 069·630·35·0 1·9 9·8 1·6
East Sutherland2·1 4·255·438·33·4 3·1 4·920·4
Gairloch3·215·550·231·08·7 4·1 1·227·5
South Ross2·2 1·284·312·43·4 5·6 1·517·5
West Grampian0·3 0·244·255·32·136·8 1·846·7
Mid-West Association1·7 1·773·223·54·0 6·3 0·329·6
South Loch Tay2·423·6 1·372·71·989·520·861·6

Among the smaller herbivores, evidence of grouse was most frequently recorded in vegetation polygons (median 29·6%). The wide interquartile range (17·5–60·3%) reflected changes in abundance across Scotland, as drier heather moor becomes more extensive and management intensifies from west to east. Hare presence (median 6·3%) appeared to have more of a bimodal distribution across DMG (around values of 3·1% and 69·1%, respectively). Like grouse, hares tended to be less frequently recorded in the west of Scotland and more frequently recorded in the central and east of Scotland. Rabbits were generally less common than hares in nine of the 11 DMG (median 2·4%), probably reflecting the fact that they prefer lower lying enclosed land with drier, mineral soils. Where Smooth grassland and heath habitats occurred close to these environments, as in some DMG in eastern Scotland, rabbit presence was recorded comparatively frequently (Table 2).

grazing and trampling impacts of different herbivore species

Averaged across all DMG, the highest predicted (median) impact was associated with the recorded presence of sheep (Fig. 2). The estimated impact of sheep was the highest of all herbivores in seven of the 11 DMG and significantly greater than zero in a further two (see Fig. S1 in the supplementary material). The second highest predicted impact was associated with the recorded presence of cattle (Fig. 2), with the impact greater than that associated with sheep in two DMG and significantly greater than zero in a further two (see Fig. S1 in the supplementary material). The third highest predicted impact was associated with the recorded presence of rabbits (Fig. 2). However, these tended to be localized effects and generally were more discernible in DMG in eastern Scotland; in particular, they tended to be associated with Smooth grassland habitats (see Appendix S1 in the supplementary material). In Angus DMG, rabbit was the only herbivore associated with a significant predicted impact averaged across all habitats. For hares, the average predicted impact associated with their recorded presence was very small (Fig. 2). However, in one DMG, East Sutherland, hares did have a significant effect, though this was smaller than either sheep or cattle (see Fig. S1 in the supplementary material). Closer inspection of the herbivore–habitat interaction terms in the fitted models suggested that in some other DMG relatively heavy hare impacts were restricted to Heath.

Figure 2.

The median (with 5–95% ranges) grazing and trampling impacts associated with the recorded presence of each herbivore species, averaged across all DMG and all habitats. Results for individual DMG are given in Fig. S1 in the supplementary material.

In contrast, the average predicted impact associated with the recorded presence of red deer and red grouse tended to be negative (Fig. 2) Thus the vegetation polygons having signs of red deer presence had predicted impact scores lower than similar vegetation polygons without signs of their presence. However, in one DMG, Cairngorm–Speyside, the recorded presence of red deer did tend to increase the predicted impact, although this was smaller than either sheep or cattle (see Fig. S1 in the supplementary material). Closer inspection of the herbivore–habitat interaction terms in the fitted model for Cairngorm–Speyside suggested that red deer impacts were significant in five of seven habitats (Fig. 3).

comparing sheep and deer impacts by habitat

The presence of sheep tended to most frequently have a significant impact on Smooth grassland habitat across Scotland (10 of 11 DMG were significantly impacted), and least frequently on Blanket bog (five of 11 DMG) (Fig. 3). As described above, the presence of deer tended to be more diffuse. Overall there was a significant impact in only seven of 63 (11·1%) habitat–DMG combinations (two on Blanket bog but none on Coarse grassland) for deer compared with 58 of 77 (73·5%) for sheep (Fig. 3). A more detailed description is given in Appendix S2 in the supplementary material.

In 40 of 63 (63·5%) habitat–DMG combinations, where both red deer and sheep impacts could be estimated, the predicted impact of the recorded presence of sheep was significantly greater than the predicted impact of deer (Fig. 3). In contrast, in only one of all the 63 (1·6%) habitat–DMG combinations, Blanket bog in Cairngorm–Speyside, was the predicted impact of the recorded presence of deer significantly greater than sheep (P = 0·045; see Table S2 in the supplementary material).

sheep and deer impacts at local stocking densities

For East Sutherland and West Grampian, the only two DMG for which both sheep and deer density estimates were available at the estate scale, non-linear (asymptotic) regression revealed that the predicted impact associated with the presence of deer increased significantly with deer density, whereas the estimated impact associated with the presence of sheep tended not to vary with sheep density (Fig. 4). Thus, even at apparently low local densities, sheep had a pronounced impact compared with low densities of deer. However, in West Grampian, where predicted impacts were generally higher than in East Sutherland, at densities over 40 animals km−2 deer had similar impacts to sheep (Fig. 4b).

Figure 4.

The mean grazing and trampling impact scores associated with sheep (open circles) and deer (closed circles) estimated for different land management units (estates) within (a) East Sutherland DMG and (b) West Grampian DMG, plotted against the density of each herbivore on each estate. The fitted asymptotic regressions for sheep and deer are significantly different in both DMG.

deer impact and density at regional scales

The distribution of grazing and trampling impact scores differed significantly between the 11 DMG areas (χ2 = 2928, d.f. = 40, P < 0·001). Impact scores tended to be lower in the DMG in the north and west of the Highlands and higher in the south and east of the Highlands (Fig. 1). In three habitats (Blanket bog, Heath and Coarse grassland), the fitted asymptotic regression indicated that deer population density explained a significant proportion of the variation in mean impact between DMG (r2 = 0·5561, 0·6891 and 0·2845, respectively; Fig. 5a–c). However, on Smooth grassland with bracken and/or rushes (Fig. 5d) and three others not shown (Smooth grassland, Montane grassland, and Montane heath), deer density did not explain the variation in estimated impact.

Figure 5.

The estimated mean grazing and trampling impact scores within each of the 11 DMG plotted against the deer density from counts by the Deer Commission for Scotland, for (a) Blanket bog (y = 0·9042 ln(x) − 0·246), (b) Heath (y = 1·1003 ln(x) + 0·0502), (c) Coarse grassland (y = 0·6284 ln(x) + 0·505) and (d) Smooth grassland with rushes and/or bracken (y = 0·558 ln(x) + 1·7683).


The rapid habitat assessment methods and subsequent analyses detected differences in the grazing and trampling impacts associated with different herbivore species at a variety of scales, ranging from land management/ownership units (10–100 km2) to subregional (deer population) scales (500–2000 km2) between areas of Scotland. At larger scales it was also possible to identify interactions between herbivores and habitats, not only for deer and sheep, which were very widely distributed across vegetation polygons, but also for those more locally restricted, as in the case of cattle, rabbits and mountain hares. The simple approach adopted here could be applied quite easily to other temperate and tropical multispecies grazing systems.

The evidence that the recorded presence of sheep was associated with higher grazing and trampling impacts than the other four mammalian herbivores was compelling. Sheep were associated with the highest impact across averaged habitats in seven out of 11 DMG, and increased the probability of observing a ‘moderate’ or greater impact in most habitats, not only those dominated by grasses but also on Heath. After sheep, the recorded presence of cattle was most commonly linked with increased impact on open-hill habitats, although their impact was localized, because cattle occurred in fewer habitats and a lower proportion of vegetation polygons within habitats. None the less, the estimated mean impact associated with cattle was slightly higher than sheep when averaged across habitats in three DMG.

In contrast to domestic stock, the recorded presence of red deer, mountain hares and rabbits had comparatively little impact on plant communities at the DMG scale; each species significantly increased the impact over all habitats in only one DMG (Cairngorm–Speyside, East Sutherland and Angus, respectively). However, in some other DMG the estimated grazing impact associated with the recorded presence of the wild mammalian herbivores was specific to particular habitat types, for example deer on Blanket bog and Montane heath in South Loch Tay, rabbits on Smooth grassland in both South Loch Tay and Cairngorm–Speyside, and hares on Heath in the Midwest.

distinguishing sheep and deer impacts

Although differences in diet selection and forage patch utilization by sheep and red deer are apparently subtle and vary with the grain of vegetation patch mosaics (Clarke, Welch & Gordon 1995; Hester & Baillie 1998; Hester et al. 1999; Palmer & Hester 2000; Palmer et al. 2003), the estimated impact associated with the recorded presence of sheep was repeatedly greater than the impact associated with deer. In no less than 40 of the 63 (63·5%) habitat–DMG area combinations where we could directly compare the two species because we had sufficient samples of recorded presence/not recorded present in habitats, the probability of sheep impact was significantly greater than deer impact. In only one habitat, Blanket bog, in one DMG area, Cairngorm–Speyside, did the estimated impact of deer grazing significantly exceed the estimated impact of sheep.

The success of our approach in distinguishing the magnitude of impacts associated with the presence of sheep vs. red deer probably reflects the fact that it assessed grazing impact at the extant scale of plant communities, thus recognizing that even 0·25-km2 mapping units may be heterogeneous, with multiple patches of two or more habitats. Although the recorded presence/not recorded present (binary) data for herbivores may have included some false zeros, the assessment of large numbers of polygons in each DMG (median 1067) usually gave sufficient patches that had not been visited recently by one or more herbivore species, enabling us to estimate the independent effect of each herbivore species. Although in two DMG (North Ross and West Grampian), with relatively low numbers of vegetation polygons, it was not possible to estimate deer impacts because of their extremely high recorded presence (> 97%) this was not a problem in the other nine DMG. Otherwise, there were only problems of robust estimates of impact for extremely low recorded presence, particularly for rabbits, where in five DMG they were recorded in less than 2% of vegetation polygons.

Unfortunately, our methods did not estimate the relative degree of use by each herbivore, where it was recorded present in a vegetation polygon, and only latterly were landowners/managers questioned about their sheep management practices and stocking rates at estate scales. However, in the two DMG where we could investigate the relationship between estimated impact of sheep vs. red deer in relation to their respective densities simultaneously, we found little or no increase in impact associated with higher densities of sheep, but significant increases in deer impact as deer densities increased. In the West Grampian DMG, the estimated impacts of deer only reached those of sheep at about 50 head km−2. However, in East Sutherland DMG, where impacts tended to be lower than West Grampian, sheep densities tended to be lower than deer densities (in 16 of 18 estates) even before taking into account that often sheep tended to be on the open-hill for only part of the year.

The apparent lack of a relationship between grazing impact associated with sheep and their density is puzzling. One possible explanation, which would be consistent with the lower recorded presence of sheep compared with deer in vegetation polygons (median 43% and 90%, respectively), is that sheep are more aggregated than red deer, so that as sheep numbers increase they occupy proportionally more range but the effective density per unit area remains similar. While sheep may be more socially gregarious, their aggregation is likely, at least in part, to be a simple consequence of the way sheep are managed, with flocks tending to be grazed in particular locations, often for restricted periods, dependent on their stocking density (Lawrence & Wood-Gush 1988). Although the ranging behaviour of both species has not been studied sympatrically, there is evidence that sheep do have smaller home ranges compared with deer (30 ha vs. > 100 ha, respectively; cf. Hewson & Wilson 1979; Clutton-Brock, Guinness & Albon 1982), and hence sheep may be more likely to remain at local, hillside scales (Lawrence & Wood-Gush 1988). An alternative, but not necessarily mutually exclusive explanation, could be that, in some areas, sheep densities had been reduced shortly before the surveys and there had been insufficient time for the vegetation to recover.

Although our models included terms representing the interactions between herbivores, these were fitted with an expected additive effect of increased impact associated with pairs of species with similar diet and habitat use. However, we did not expect to find lower recorded impacts on vegetation polygons containing signs of herbivore presence than on similar habitats without signs of that herbivore, as in the case of red deer in five of the seven habitats in East Sutherland. Although herbivores can drive vegetation state transition and influence productivity (van der Wal 2006), we have no evidence in this case that deer were having a beneficial effect within habitats. Nor was there any evidence that they avoided sheep more in this DMG than others (55% deer-only polygons in East Sutherland was the median across DMG). While it could be a chance effect, given the complexity of the fitted model, it is possible that deer sought out patches within habitats that were particularly productive because of edaphic conditions, and in effect the DMG appeared lightly grazed.

management implications

Our results have important implications for managers of rangelands in Scotland. First, our models offer the potential to produce interpolated maps of predicted grazing and trampling impacts, as a function of herbivore presence/absence, vegetation and other biotic and abiotic measures (Brewer et al. 2004). However, the maps would not predict impacts relative to changes in number or density of herbivores other than deer, as the impact associated with the presence of sheep was not related to their density, and for the other herbivore species density was not estimated. For red deer, we can predict the probable average impact of either reducing or increasing numbers across habitats at the estate scale (10–100 km2) and within habitats at the regional scale (500–2000 km2).

Secondly, our results can help conservation managers anticipate the possible effects of changes in grazing pressure on the natural heritage (Thompson et al. 1995). It is likely that the total sheep stock will continue to decline, because recent reform of the Common Agricultural Policy (CAP) has moved into a post-productionist phase where good environmental stewardship is rewarded. Where sheep are removed from an area, the reduction in grazing and trampling pressure should halt further degradation of heather-dominated communities (Armstrong & Milne 1995). However, as red deer also prefer to forage in grass patches, they may fill the ‘vacuum’ left by the removal of sheep, hindering recolonization by heather (Hope et al. 1996). Unfortunately, understanding of the extent to which red deer will change their foraging and ranging behaviour when sheep are removed is largely anecdotal (but see Clutton-Brock & Albon 1992). Addressing this knowledge gap is important, particularly given that the numbers of red deer are likely to continue to grow because of higher recruitment associated with increasingly mild winters and earlier onset of spring (Albon & Clutton-Brock 1988). In Heath habitats, deer densities above about 15 deer km−2 were associated with impacts of moderate or higher (Fig. 5), thus pro-active deer management to constrain the rate of deer population growth may be necessary if the priority is to halt the loss of heather.

Thirdly, there is an opportunity to reconsider the issue of whether heavy herbivore impact, leading to alternative plant communities, is an undesirable outcome rather than a dynamic process between alternative stables states (van der Wal 2006). Light or light/moderate grazing will maintain heather-dominated habitats but grass-dominated ones require moderate or greater levels of grazing. The debate is really about how herbivores are managed to create or maintain landscapes with different ecological properties and visual characteristics: the shape of which will depend on land managers aims in meeting either public and/or private objectives.

future research

While our approach was capable of distinguishing the grazing impact of different herbivores on open-hill habitats, this was scaled in terms of the relative condition of the dominant species within a vegetation community, and not in terms of the implications for habitat condition, plant diversity/species richness or wider biodiversity consequences. The implications of different grazing and trampling impact classes will vary between communities. For example, moderate grazing impact may actually increase species richness, by reducing the cover of dominant species and enhancing the competitive ability of other species (Clutton-Brock & Ball 1987; Gordon 1988; Welch & Scott 1995; Virtanen, Edwards & Crawley 2002). Furthermore, changes in physical habitat structure associated with cattle grazing may have cascading effects beyond the plant species composition, including increased invertebrate abundance benefiting insectivorous birds, and higher abundance of voles, compared with similar levels of grazing with sheep only (Evans et al. 2006). However, there are concerns that the facilitation of small selective herbivores, such as voles, by larger species, such as cattle, may reduce plant diversity because the smaller herbivores prefer rare, palatable species (Olff & Ritchie 1998).

Unfortunately the evidence from individual studies around the world indicates that the magnitude and direction of the effects of different herbivores varies over spatial and temporal scales. Such complex interactions present a challenge to land managers and conservationists alike, and have led to the emergence of new conceptual frameworks describing the influence of herbivores on plant diversity across gradients of soil fertility and rainfall (Van de Koppel et al. 1996; Olff & Ritchie 1998). These should be tested at a range of scales, in order to provide further insights into the adaptive management of herbivore–habitat interactions to enhance biodiversity (Gordon, Hester & Festa-Bianchet 2004).


This synthesis was funded by the Scottish Executive Environment and Rural Affairs Department (SEERAD) through the discretionary funding available to the directors of SEERAD-sponsored institutes. The development and testing of the sampling methodology and surveys in 1997–99 were co-funded by SEERAD, the Deer Commission for Scotland and Scottish Natural Heritage (SNH). Surveys from 2000 onwards were commissioned by individual DMG, part-funded by SNH. In particular, we would like to thank the DMG and estate owners, managers and stalkers for the provision of information, and assistance with access, to enable these studies to be carried out. We are grateful to all the field survey team members, especially Anneke Stolte, David Henderson and Clare Waterhouse, and Andrew Dalziel for assembling the overall database. Access to the Beowulf Linux computer cluster, Rowett Research Institute, Aberdeen, was assisted by Tony Travis. The manuscript benefited from advice on the analysis, arguments, structure and figures given by Mick Crawley, David Elston, Maggie Gill, Alison Hester, Glenn Iason, John Milne, Josephine Pemberton, Simon Thirgood, Des Thompson and Rene van der Wal. We are grateful to them all, as well as the Associate Editor, E. J. Milner-Gulland and two anonymous referees, for their constructive criticisms.