Grazing mammals are often used to maintain and restore high conservation value plant communities, but the evidence base for management is lacking long-term studies.
We erected grazing exclosures in dune slacks to determine the impact of three different grazing regimes on the plant community: (1) rabbits and sheep excluded for 36 years, (2) continued rabbit grazing for 36 years and (3) rabbit grazing for 17 years followed by rabbit and sheep grazing for 19 years. We monitored plant community composition inside and outside the exclosures.
All of the plant communities changed over time, moving away from the original high-value system and losing some characteristic species. Grazing slowed succession, reduced woody perennial cover and increased graminoid and forb cover and species diversity. The impact of adding sheep grazing to the existing rabbit grazing was additive at the functional group scale, but both complementary and additive (depending on the species) at the plant species scale.
Synthesis and applications. At the levels of grazing present in this study (2·5 sheep ha−1 year−1), sheep had similar impacts on dune slack plant communities to rabbits, making them suitable for replacing or augmenting rabbit grazing for conservation management. At the intensity present in this study, long-term grazing can help to maintain a species-rich dune slack community but is not sufficient for successful restoration.
Grazing mammals have large impacts on natural and semi-natural ecosystems and can exert strong top-down control, affecting the structure, composition and productivity of plant communities (Huntly 1991; Michunas & Lauenroth 1993; Olff & Ritchie 1998). The magnitude of these impacts and the widespread distribution of mammalian herbivores mean that understanding plant community responses to mammalian herbivory is a problem of global significance. However, plant community responses are complex and can be difficult to predict, making management decisions difficult.
Herbivore impacts on plant community composition can be highly dependent on herbivore species identity. Species can differ in the intensity and selectivity of grazing and in their digestive efficiency (Danell & Bergström 2002). Herbivore body size can also be an important determinant of their impacts on plant communities (Olff & Ritchie 1998; Bakker et al. 2006). For example, intermediate-sized digging mammals might differ from large domestic herbivores because the burrows they dig to escape predation create infrequent, intense disturbance which can increase diversity (Huntley & Reichman 1994). Conversely, large herbivores create frequent, less intense disturbance through trampling (Olff & Ritchie 1998). In addition, smaller herbivores might be more selective than large herbivores because they forage at smaller spatial scales (Laca et al. 2010). Herbivore impacts can also interact due to these differing impacts on the plant community and their impacts on the other species behaviours (DeGabriel et al. 2011). These interactions can have additive (i.e. similar impacts, intensifying plant community change) or compensatory (i.e. different impacts moderating plant community change) effects (Ritchie & Olff 1999). In addition, the short- and long-term effects of herbivores can be very different (e.g. Olofsson, de Mazancourt & Crawley 2007). Therefore, careful species selection for conservation grazing is important and must consider the impacts of individual species, interactions between the domestic herbivore and any existing ‘wild’ herbivores, and long-term as well as short-term impacts.
In this study, we monitored the impact of different grazing regimes on dune slack plant communities. These low-lying seasonally flooded areas are a component of coastal dune systems and are of high conservation importance (e.g. being listed in Annex 1 of the EC Habitats and Species Directive (Council Directive 92/43/EEC)). Coastal sand dunes are highly dynamic systems and classic models for short-term (i.e. <100 years) vegetation changes during primary succession on land (Walker et al. 2010). Dune slack plant communities are created by plant establishment after severe, infrequent disturbance. High conservation value plant communities develop after an early pioneer stage (Grootjans et al. 2008). In the absence of grazing or further disturbance, dune slack plant communities follow a clear succession towards scrub (Lammerts et al. 1999; Van der Hagena 2008). As such, stabilization and reductions in grazing intensity (e.g. due to myxomatosis–Armour & Thompson 1955; Hudson, Thompson & Mansi 1955) pose a threat to these habitats (Ranwell 1960; Grubb & Suter 1971; Sheail 1991). Grazing management is a potentially useful tool in helping to maintain high plant diversity in coastal dune systems by creating disturbance and preventing scrub and woodland establishment. However, there are few long-term studies of the impact of grazing on these high-value systems.
Despite their relatively recent introduction into many systems, the European wild rabbit Oryctolagus cuniculus can have large-scale (positive and negative) impacts on plant communities (e.g. Zeevalking & Fresco 1974; Delibes-Mateos et al. 2008; Lees & Bell 2008). Rabbits have been introduced onto every continent except Antarctica and in some systems have replaced native herbivores. They are therefore a globally important ‘wild’ herbivore. However, rabbit populations can be variable, unpredictable, and their grazing habit may not necessarily be the most appropriate for optimum conservation outcomes. As such, domestic herbivores are now commonly used to augment rabbit grazing in order to maintain species diversity on dunes (Hewett 1985; Kohyani et al. 2008) and other habitats (e.g. chalk downland–Sheail 1991). Domestic herbivores are often not new introductions to these various systems but reintroductions, having previously grazed (often for millennia) prior to recent removal. However, these recent reintroductions are often into a different ecological context due to human impacts on the ecosystem. Appropriate decision-making is hampered by the lack of an evidence base to support management of grazing. In particular, there is a lack of long-term studies and a lack of comparison between maintaining existing rabbit grazing or augmentation with domestic herbivores.
We established herbivore exclosures within dune slacks in Ainsdale Sand Dunes National Nature Reserve in North West England to assess the effects of three different grazing regimes: (1) rabbits and sheep excluded for 36 years, (2) continued rabbit grazing for 36 years and (3) rabbit grazing for 17 years followed by rabbit plus sheep grazing for 19 years on the plant community. We assessed changes in plant community composition over the 36-year period. The aim of this study was to investigate the role of grazing in facilitating the persistence of high conservation value plant communities. Specifically, we addressed the following questions: (1) does grazing retard succession and therefore maintain high conservation value plant communities and (2) how does adding sheep to existing rabbit grazing impact on vegetation composition?
Materials and methods
Sefton Coast Special Area of Conservation (SAC) is located in North West England and consists of a 2074-ha coastal belt of frontal dunes approximately 30 km long and 2–4 km wide. There is a transition from active dynamic dunes on the seaward side to largely stable dunes inland. Ainsdale Sand Dunes National Nature Reserve (53º35′N, 03º05′W) was established in 1965 and forms the central section of the Sefton Coast SAC. The site contains a large area of fixed dunes and dune slacks. The water-table is intensively monitored and rises and falls by over 50 cm in most years (see Fig. S1 in Supporting Information). In a typical year, 30% of the slacks are flooded to a depth of 10–30 cm, with 10% remaining flooded in the summer (Clarke & Ayutthaya 2010). Rabbits have been present since at least the late 17th century (Houston 2004) when the dunes were managed as warrens. The site still has a large rabbit population, which until 1991 was the only source of grazing. In 1991, Herdwick sheep were introduced onto the site at an approximate density of 2·5 sheep ha−1 year−1 and graze between September/October and May/June (approximately 200 sheep on 54 ha for 8 months of the year). They are moved around four enclosures and so each enclosure is grazed intermittently throughout the winter. During the study, mean ± SE annual rainfall was 849 ± 18·4 mm; mean annual maximum and minimum temperatures were respectively 13·6 ± 0·2 °C and 6·1 ± 0·1 °C. N deposition was approximately 13·7 ± 0·3 kg N ha−1 year−1 and remained relatively constant throughout the study (see Fig. S2 Supporting Information). This level of N deposition is intermediate for the UK and is at the critical load threshold for this habitat of 10–15 kg N ha−1 year−1 (Bobbink & Hettelingh 2011).
In 1974, we erected rabbit exclosures in four dune slacks. These slacks were located along a 1-km transect orientated parallel to, and approximately 500 m from, the coast line, with one slack at either end and two located 60 m apart in the centre. Within each slack, three 1-m-high, 1·5 × 1·5 m rabbit exclosures were constructed. In 1991, when sheep grazing was introduced, sheep exclosures (1-m-high, 10 × 20 m) were erected around the rabbit exclosures at two of the dune slacks. These exclosures were impermeable to sheep, but permeable to rabbits. At the same time, new rabbit exclosures were erected in the same locations as, and to the same specification of, the original exclosures. The rabbit exclosures were also successful in excluding sheep from the plots.
Initially, vegetation was surveyed in one 1 × 0·5 m plot within each rabbit exclosure (n =3 per slack) and two 1 × 0·5 m plots per slack outside of the exclosures (with the exception of one dune slack where 3 such plots were established). Where erected, the sheep exclosures enclosed all survey plots in the dune slack. Therefore, additional survey plots were established outside of the sheep exclosures in 2009 (n =3 per slack). The numbers of plots and the grazing regime for each plot are shown in Table 1. Vegetation composition was measured immediately after the rabbit enclosures were erected in July 1974 and then in June/July of 1975, 1976, 1984, 1985, 1986 and 2009. The exceptions were the plots established in 2009 (only measured in 2009) and plots where the rabbit exclosures disintegrated and therefore failed to exclude rabbits or sheep (not measured in 2009).
Table 1. Details of grazing treatments throughout the duration of the study
Table 2. Variance attributed to time and grazing treatment (including treatment × time interaction) in the principal response curves (PRC) for the species and functional group data sets. Also presented is the percentage of treatment variance that is explained by the first axis
Axis 1 (%)
Table 3. Univariate linear mixed model (LMM) results for impacts on dune slack plant communities growing with different grazing regimes. Presented are the P-values from the LMM results for grazing regime (grazing removed in 1974 or continued rabbit grazing), survey date and their interaction for the period 1974–1986; and the effect of grazing regime (grazing removed in 1974, continued rabbit grazing or sheep grazing added to rabbit grazing in 1990) on measures in 2010
Grazing regime X Time
Number of hits
Vascular plant spp. richness
Non-vascular plant spp. richness
Simpson's diversity index: higher plants (1/D)
Number of faecal rabbit pellets.
Salix repens height
Vegetation composition was measured using the line-point intercept method. Wire pins (2 mm in diameter) were inserted vertically into the vegetation at 5-cm intervals along a 1-m V-shaped transect (n =20 pins per plot). The number of times each plant species touched each pin was recorded. The number of hits per species was used as a measure of the absolute (total number of hits for species) and relative (number of hits for species/total number of hits for all species in plot) abundance of that species. This method gives a good objective, quantitative assessment of species abundance (Godínez-Alvarez et al. 2009). We also put vascular species into four broad functional groups based on those identified by Box (1996), using physiognomic characteristics. The functional groups used were forbs, graminoids, woody perennials and pteridophytes. The height of Salix repens in each plot was also measured at 10 random points in each plot.
Pellet counts can provide a good approximation of relative rabbit abundance (Palomares 2001). Therefore, to quantify rabbit abundance on the plots, we counted the number of faecal pellets within each 1 × 0·5 m plot in 2009, at the same time as the vegetation surveys.
Ellenberg values were used to indicate the plant community response to changing environmental conditions resulting from the different grazing management or changes over time. Ellenberg values are indicators of environmental conditions, based on the propensity of species to grow within specific environmental limits. Thus, the presence of a species indicates the prevailing environmental conditions (Ellenberg 1988). Ellenberg values for all native UK species have been determined by Hill et al. (1999). We used the Ellenberg values for each species, taken from Hill et al. (1999), to calculate a mean abundance-weighted Ellenberg value for light (L) and nitrogen (N) and water (F) for each plot at each monitoring date as follows:
where A is the percentage abundance of each species and E is the Ellenberg value for that species taken from Hill et al. (1999). Simpson's diversity index (D) was calculated as follows:
where n is the percentage abundance of each species. From this Simpson's reciprocal index (1/D) was calculated. The similarity of the plant community in each plot at each sampling date to the plant community in that plot at the start of the experiment was measured by calculating the Bray–Curtis dissimilarity index for each plot at each sampling date using the vegetation analysis software: juice 7·0 (Tichý 2002). The Bray–Curtis dissimilarity index describes the comparison of species composition between two plots as a value between 0 and 1, where 0 indicates identical species composition and 1 indicates no shared species.
To explore the impact on species abundance of the grazing treatments and changes in time, we first used detrended correspondence analysis (DCA) in CANOCO 4·54 (Ter Braak & Šmilauer 2006). DCA is a method of indirect gradient analysis, which is suitable for species that display unimodal responses; this was indicated by initial exploratory analysis. For the analyses, the number of hits was converted into percentage abundance. The treatments and year of measurement were plotted passively (as ‘supplementary variables’) on the ordination to facilitate the interpretation of the data. To interpret this ordination, we considered the proximity of a species to a treatment/date on the ordination as an indication of the importance of that species for differentiating the plant community in that treatment/date from those in the other treatments/dates.
We then used principal response curves (PRCs) to investigate the impact of the grazing treatments on the plant community over time. PRCs are based on redundancy analysis (RDA) and plot the principal component of the time-dependent treatment effect against time. This produces a plot of the community response to the treatments over time (Van Den Brink & Ter Braak 1999). This approach is described in detail by Lepš & Šmilauer (2003) and was used by Pakeman (2004) to assess changes over time in the response of plant functional traits to grazing. We presented treatment effects in reference to the rabbit grazing treatment because this represents the ongoing management of the site. We also present the ‘species’ weights, which show ‘species’ abundance changes in the treatments relative to the reference. A weighting at or around zero indicates no impact on the treatment effect (the ‘species’ being either ubiquitous or showing no consistent change in abundance with treatment). We carried out a separate PRC analyses by species abundance and by plant functional group abundance with weighted Ellenberg values.
Differences between grazing treatments and changes over time were tested in SPSS Statistics 19·0 (IBM Corp 2010) using a linear mixed model (LMM) with residual maximum-likelihood (REML) estimation for weighted Ellenberg values, species richness, 1/D, Bray–Curtis dissimilarity index, the proportional abundance of each functional group, the total number of hits and the number of faecal rabbit pellets per plot. The analysis of data from 1974 to 1986 was conducted separately to that of 2009 due to the change in experimental design with the introduction of sheep grazing in 1991. Temporal autocorrelation within plots was accounted for in the 1974–1986 data by modelling the covariance between sampling dates within plots using a compound symmetry covariance structure. For both sets of analyses, grazing treatment and the identity of the dune slack were treated as fixed factors. To determine the value of including slack identity as a blocking factor, the suitability of the model was tested using Akaike Information Criterion (AIC). For the 2009 data, the difference between treatments was tested using custom hypothesis tests. We tested for differences between ungrazed and grazed plots (the mean of rabbit- and rabbit+sheep-grazed plots) and for differences between rabbit-grazed and rabbit+sheep-grazed plots.
A total of 59 vascular (18 graminoids, 35 forbs, 3 woody and 3 pteridophytes) and 15 non-vascular plant species were recorded throughout the study period (see Table S2 in Supporting Information). At the start of the study, the plots were dominated by S. repens and Agrostis stolonifera with abundant Carex flacca, Lotus corniculatus and Hydrocotyle vulgaris.
The DCA ordination showed a high degree of variation in species composition between the plots (Fig. 1, the gradient lengths for the first four axes were 3·838, 2·950, 3·166 and 2·542). The eigenvalues were 0·489, 0·373, 0·268 and 0·236, and they explained 9·9%, 7·6%, 5·4% and 4·8% of the variation in species composition, respectively. Axis 1 is dominated by differences in the ungrazed plots between 1974 and 2009, while axis 2 combines the change in rabbit+sheep and rabbit grazing between 1974 and 2009 and changes in the ungrazed plots from 1974 to 1986.
The DCA shows that the plant communities in all of the grazing treatments changed over time (Fig. 1). The largest changes were in the ungrazed plots. When grazing was removed, these plots showed a clear trajectory of changing species composition. After 36 years of grazing removal, the vascular plant community was dominated by Rubus caesius and S. repens, with the ground layer dominated by Lotus pedunculatus and H. vulgaris. As a result, these plant communities were more dissimilar to their starting point (1974) in 2009 than those in the grazed plots (Fig. 2d). Many of the measured variables changed over time and were significantly affected by grazer exclusion (Table 3). In comparison with the grazed plots, grazing removal also resulted in significantly higher woody plant abundance (Fig. 3a) and weighted Ellenberg N scores (Fig. 4b) and lower forb abundance (though forb abundance recovered in 2009) (Fig. 3b), graminoid abundance (Fig. 3c) and species diversity (Fig 2b & c).
The plant community in the grazed plots followed the trajectory of the ungrazed plots, but to a lesser extent (Fig. 1). After 36 years, the plant community in grazed plots was dominated by S. repens, Festuca rubra and L. pedunculatus with H. vulgaris also dominating in rabbit+sheep-grazed plots. Also abundant in grazed (rabbit and rabbit+sheep) plots, but not the ungrazed plots, were Carex nigra, C. flacca, Anagallis tenella and A. stolonifera with Equisetum variegatum and L. corniculatus also abundant in rabbit+sheep-grazed plots. In comparison with the ungrazed plots, the composition of plant functional groups in the grazed plots was more similar to that at the start of the study and species richness was significantly higher (Table 3, Fig. 2d). In addition, the increase in woody plants cover seen in the ungrazed plots was prevented (Table 3, Fig. 3a). There were no statistically significant differences in any of the measured variables between rabbit- and rabbit+sheep-grazed plots (Table 3, Figs 2-4). However, for all measured variables (with the exception of Ellenberg values), the impact of adding sheep grazing consistently enhanced the effect of grazing. For example, species richness (Fig. 2b), graminoid cover (Fig. 3c) and forb cover (Fig. 3b) were highest for rabbit+sheep-grazed plots; woody plant cover, Bray–Curtis dissimilarity (Fig. 2d) and non-vascular plant cover were lowest for rabbit+sheep-grazed plots (Fig. 3d).
Salix repens was present in every plot, and plants were taller in the ungrazed treatments than in either of the grazed treatments (Table 3; mean height ± SE = ungrazed: 44·1 ± 4·2 cm, rabbit grazed: 19·5 ± 6·0 cm, rabbit + sheep grazed: 19·7 ± 3·6 cm).
The principle response curve ordinations explained 30·55% and 32·98% of the variation for species and functional groups, respectively (Table 2, sum of time and treatment scores). The proportion of variation accounted for by time was very similar to that accounted for by treatment (including the treatment × time interaction). Axis 1 accounted for 49·1% and 55·2% of the treatment variation for species and functional groups, respectively.
The PRCs (Fig. 5) confirm that after rabbits were excluded, the plant community in the ungrazed plots changed in comparison with those plots that continued to receive the existing management of allowing rabbit grazing. This difference was apparent in both the species composition (Fig. 5a) and the functional groups (Fig. 5b). The addition of sheep grazing in 1991 also resulted in changes to the plant community, but in the opposite direction to the ungrazed plots. The abundance of A. stolonifera exhibited the strongest response to grazing treatment, decreasing when grazing was removed and increasing when sheep grazing was added. C. flacca, Mentha aquatica, H. vulgaris, A. tenella and Prunella vulgaris all showed the same response, but to a lesser extent, and Trifolium repens, C. nigra, Juncus articulatus, F. rubra and Bryum spp. to an even lesser extent. Poa pratensis was the species exhibiting the strongest positive response to grazing removal, followed by R. caesius, Hippophae rhamnoides and Eurynchium praelongum. The grazing treatment impact was strongest for forbs and graminoids, which decreased when grazing was removed and increased when sheep grazing was added. Woody species and those with high Ellenberg N and F values also had a strong response to grazing, increasing when grazing was removed and decreasing when sheep grazing was added.
No faecal rabbit pellets were found in the plots inside the rabbit exclosures, and there was no significant difference in the number of faecal rabbit pellets found in the plots inside and outside the sheep exclosures (Table 3).
Dune slack plant communities are valued, biodiverse systems. Forbs, graminoids and non-vascular plants were the most species-rich functional groups in the dune slacks studied, containing considerably more species than woody plants and pteridophytes. The removal of rabbit grazing resulted in an increase in the abundance and height of woody plants, a decrease in the abundance of graminoids and forbs and a loss of diversity. As such, in terms of biodiversity in the dune slacks studied here, the primary benefit of grazing is in the maintenance of a low abundance of woody plants, a short sward and high abundance of the more species-rich functional groups of forbs and graminoids. This maintenance of low abundance and height of woody plants was sufficient to significantly retard succession in the dune slacks. As such, maintaining rabbit grazing and introducing grazing by sheep at a stocking density of 2·5 sheep ha−1 year−1 slowed the rate and altered the trajectory of vegetation change and maintained species diversity.
In our study, when grazed by rabbits, the plant community initially (in the first 12 years) retained characteristic species and increased in diversity, in contrast to the plots where rabbits were excluded. Similar positive impacts of grazing have been shown in dune plant communities (Plassmann, Edwards-Jones & Jones 2009) and are common in other temperate systems (e.g. Olff & Ritchie 1998). However, after 36 years, the grazed plant community had changed significantly from the original community, with the grazing-tolerant graminoid F. rubra becoming co-dominant in plots grazed by rabbits and those where sheep were also introduced. In addition, the species of the bryophyte in the genus Bryum, which is characteristic of dune slack vegetation, was absent from all plots. This may indicate that the plant community is developing away from that of an early successional community into that of a grazed system. In the case of the management of dune slacks, this is undesirable. However, after 36 years, the grazed plant communities still retained characteristics of dune slack communities, dominated by forbs and graminoids, whereas the ungrazed plots had transitioned to scrub, dominated by R. caesius and S. repens. This positive impact of grazing indicates that it is a valuable management tool, even in cases where grazing alone might not be sufficient to maintain the desirable early successional communities.
How Does Adding Sheep to Existing Rabbit Grazing Impact on the Ecosystem Response?
In our study system, it was only possible to include sheep grazing with rabbit grazing because it was impractical to include sheep and exclude rabbits. Our results must therefore be interpreted in this context. In 2009, rabbit abundance in the rabbit-grazed and rabbit + sheep-grazed plots did not differ, based on faecal pellet counts. Although based on only 1 year of data, this does indicate no difference in rabbit grazing pressure between grazed plots. However, the rabbit population may have changed over time, and adding sheep in 1991 could have impacted rabbit populations across all plots. These differences would not be reflected in pellet counts in 2009. Nonetheless, while the addition of sheep grazing is likely to be confounded with grazing intensity, this reflects the reality of management at the study site.
The impact of the two herbivores on the dune slack plant community was very similar when comparing changes in the abundance of plant functional groups. At this scale, the addition of sheep had an additive impact–that is, a greater, but qualitatively similar impact. However, when considering plant community composition at a species level, there were some subtle differences. For example, for some species, the addition of sheep grazing had an additive effect (e.g. the abundance of F. rubra, L. corniculatus and E. variegatum was higher in rabbit-grazed plots than in ungrazed plots, but was even higher in plots grazed by both rabbits and sheep). Conversely, for H. vulgaris, the addition of sheep grazing had a complimentary effect [i.e. abundance was lower in rabbit-grazed plots (1·25 hits per 20 pins) than in ungrazed plots (6·00 hits per 20 pins), but considerably higher in plots grazed by both sheep and rabbits (16·09 hits per 20 pins)].
Ritchie & Olff (1999) showed the importance of considering additive vs. complementary impacts of herbivores when different species graze together. Our results suggest that these two mechanisms might both operate in a single system, but at different scales. This might be because different herbivores can impact the system in the same way (e.g. similar selectivity and intensity of grazing), but in other respects their impacts can differ (e.g. different disturbance effects). These differences could be explained by body size, which can have particularly important influences on their grazing impacts (Olff & Ritchie 1998; Bakker et al. 2006). For example, sheep are larger so would create different patterns of disturbance to rabbits. If this is the case, the additional disturbance caused by sheep might benefit H. vulgaris, L. corniculatus and E. variegatum, possibly by facilitating colonization. This has important consequences for conservation management. H. vulgaris and E. variegatum are both characteristic dune slack species (Rodwell 2000), with E. variegatum being in general restricted to only dune slacks. The increase in the abundance of these key species might therefore indicate an important conservation benefit of adding sheep grazing to the management system.
In conclusion, the evidence from this study suggests that in the long term (36 years), grazing was not successful at preventing succession; though, grazing did retard succession in a valuable way. When herbivore grazing was removed, there was a clear succession towards a woody perennial–dominated plant community (due to the invasion of R. caesius and increased height of S. repens). This succession was associated with decreased plant species diversity and increased above-ground biomass (indicated by increased pin intercepts). The grazed plant communities changed less over time but did follow this successional trajectory to some extent. The additive effect of sheep is an indication that they may be suitable for replacing or augmenting rabbit grazing. However, at a species level, the plant communities were subtly but significantly altered when sheep grazing was added. The implication is that in this dune slack system, interactions between herbivores at this level might have had little impact on functional group composition, but a significant impact on species composition. Finally, it appears that increasing grazing pressure by introducing domestic herbivores was not sufficient to maintain the plant communities in a desirable state and the re-establishment of disturbance followed by grazing might be a required management strategy. The amount and timing of grazing must also be considered, and it might be the case that different herbivore densities would have altered the response of the vegetation.
This study was supported by funding to JM from the Botanical Society of the British Isles. Many thanks to Natural England for hosting the experiment and allowing continued access. The manuscript was significantly improved through the suggestions of Brian Wilsey and two anonymous referees.