1. The recreation of species-rich grassland represents a key EU agri-environment policy initiative intended to maintain native biodiversity and to support the provision of ecosystem services. Understanding the long-term potential for recreation success is crucial to the evaluation of such schemes.
2. We use a single-site long-term data set (22 years) to test the consequences of grazing recreation management in re-establishing plant community composition and functional trait structure as assessed relative to pristine examples of target floodplain meadows.
3. Following a July hay cut, late summer grazing of the re-growth by either sheep or cattle resulted in an increase in the similarity of plants species composition to the target floodplain meadows, but only in terms of what species had colonized, not in terms of their relative frequencies.
4. Where grazing in late summer was applied, the functional traits of the meadows undergoing recreation became similar to those of the target floodplain meadows only where grazing management was used. When plant traits were divided into subcategories (e.g. regeneration, seed biology, life-form, environmental associations), only those traits linked with plant phenology failed to show evidence of a temporal shift towards the functional trait structure of floodplain meadows.
5.Synthesis and applications:. Under typical grazing management colonization by the majority of species that characterize the target habitat type is predicted to take over 150 years. In contrast, recreation of functional trait structure can occur over a considerably shorter time-scale (>70 years). The potential to provide functionally equivalent grasslands that deliver analogous ecosystem services to those of the target habitat type is therefore a more realistic goal for recreation. We suggest that the time-scale needed to recreate grasslands puts into question the benefits of compensation schemes that allow grasslands to be lost to development (i.e. gravel extraction) in exchange for future recreation at other sites.
The quantification of how successful recreation has been (i.e. whether the re-establishment of a specific community type has occurred) is of fundamental importance to both conservationists and policy makers alike (Fagan et al. 2008; Matthews & Spyreas 2010). Recreation of particular communities with specific combinations of species has often proved hard to achieve (Bakker & Berendse 1999; Pywell et al. 2002; Edwards et al. 2007). However, if plant communities are considered in terms of functional traits, then while species establishment may differ between recreation sites, a predictable succession may still occur in terms of the functional trait structure of the community (Wilson 1999; Temperton et al. 2004; Fukami et al. 2005). It is important to identify whether the long-term recreation of grassland communities results in the re-establishment of species characteristic of a target community, or whether success can be achieved only in terms of the recreation of functional attributes. This has implications for how policy makers measure success and as such value financial incentives to promote recreation (Matthews & Spyreas 2010; Posthumus et al. 2010).
Here, we assess this issue using a 22-year data set from a floodplain meadow recreation experiment. Floodplain meadows are a habitat of conservation importance at both UK and European scales (Gowing et al. 2002) and cover as little as 1500 ha in England and Wales (Jackson & McLeod 2000). In addition, the recreation of species-rich hay meadows represents an important policy aim of the UK Biodiversity Action Plan (2006). We compare temporal changes in both plant species composition and functional trait structure relative to target examples of surviving floodplain meadow communities. We predict that the recreation of functional trait structure of grasslands will occur at a faster rate than that of species composition. It is only by understanding the long-term successional dynamics of grassland recreation that we can assess whether or not the re-establishment of degraded or destroyed plant communities can realistically be achieved (Matthews & Spyreas 2010).
Materials and methods
In the UK, floodplain meadows (defined as the MG4 Alopecurus pratensis- Sanguisorba officinalis community in the British National Vegetation Classification; Rodwell 1992) are species-rich grasslands, diverse in both grasses, dicotyledonous herbs and invertebrates (Rodwell 1992; Woodcock et al. 2006). While once widespread in lowland England and Wales, conversion to other land uses, changes in water table levels and intensification of existing management practices have resulted in widespread declines (Gowing et al. 2002). Recreation was attempted at Somerford Mead (Latitude 51°47′04″N; Longitude 1°18′43″W) at the University of Oxford’s Field Station, Wytham, UK. This was a floodplain meadow that was converted to arable agriculture in 1982 following an extended period of improvement using inorganic fertilizers. As high soil fertility (in particularly phosphorous) may limit recreation success (Janssens et al. 1998), a crop of barley was grown without inorganic fertilizer in 1985. Following this, in 1986, a 3·6-ha area was re-seeded with seeds harvested from a local floodplain meadow (Oxey Mead, Latitude 51°47′34″N; Longitude 1°18′20″W) using a brush harvester (McDonald 1992). After broadcasting these seeds, the field was harrowed and rolled to ensure the seeds were well covered. In 1987, traditional management was applied to the whole site, based on a hay cut in the first week of July, followed by aftermath grazing by sheep and cattle (Gowing et al. 2002). In October 1989, the site was divided into nine c. 0·4 ha experimental plots, separated by temporary electric fencing. Three recreation treatments (each with three replicates) were randomly allocated to these nine experimental plots. Each treatment received the July hay cut, but differed in subsequent aftermath grazing management. The three treatments were as follows: (i) control, with no grazing following the hay cut; (ii) cattle grazing, using two Friesians cows in each plot and (iii) sheep grazing, based on ten mule ewes in each plot. In all cases, grazing occurred yearly in October until the sward reached a height of c. 5 cm as assessed using a sward stick (approximately 4–5 weeks) (Stewart, Bourn & Thomas 2001). Nutrient deposition from flooding and from livestock manure would have occurred, although no inorganic fertilizers were applied (McDonald 1992; Rodwell 1992; Gowing et al. 2002). Between 1986 and 1990, the soil phosphorous dropped from 10–15 mg L−1 to 0–9 mg L−1. There are unfortunately no long-term data on soil nutrient status.
Plant community sampling
Annually from 1987 to 1989 (before the grazing experimental design had been imposed), sixty 1 × 1 m quadrats were randomly positioned across the whole experimental site. Each quadrat was divided into nine equal subunits (3 × 3 grid), and the presence or absence of vascular plant species rooted within each of these subunits was recorded. For each of the vascular plant, the proportion of the 90 subunits (based on 10 quadrats, each divided into nine subunits) within which a particular species was rooted was calculated. This is referred to hereafter as rooted frequency. These data were not used directly in any subsequent analyses as they fell outside of the replicated experimental design, but they are, however, included as reference points in the presented graphs.
From 1991 onwards, a centrally positioned permanent 10 × 10 m sampling area was established in each of the nine experimental plots. Ten 1 m2 quadrats were placed in each sampling area. A pair of quadrats were placed at the corners of each sampling area, and another pair at the centre. Each pair was separated by 1 m. Each quadrat was divided into 16 equal squares (4 × 4 grid), and the presence or absence of all vascular plant species was recorded. Similar to the process described earlier, the proportion of the 160 subunits (based on 10 quadrats, each divided into 16 subunits) that had plants rooted within them was calculated to provide a measure of rooted frequency. Patterns of change within the plant community of these nine experimental plots were monitored annually in June from 1991 to 2009, with the exception of 2000.
This same ten quadrat approach (with each quadrat divided into 16 sample squares) was used to record the rooted frequency of all vascular plants in pristine areas of existing species-rich floodplain meadows. This was undertaken at: (i) Oxey Mead in 1984, 2003 and 2004; (ii) Pixey Mead in 2003 (Latitude 51°47′12″N; Longitude 1°18′04″W); (iii) Yarnton Mead in 2003 and 2004 (Latitude 51°47′30″N; Longitude 1°18′47″W). These sites represent local examples of the target plant community. The vascular plant communities of these sites were not identical, although they had a mean between site Jacquard’s similarity of 0·72 (where 0 = no similarity and 1 = identical).
Following Reich et al. (2003), traits represent species characteristics that have evolved in response to competitive interactions and abiotic environmental conditions and were defined to be any attribute that will influence a plant’s establishment, survival or fitness. This can include both physical adaptations in terms of structure, growth form and capacity to utilize resources, as well as aspects of the timing of phenological development and establishment characteristics (Grime, Hodgson & Hunt 1988; Pywell et al. 2003; Reich et al. 2003). As plants show a large number of both correlations and trade-offs between traits, typically as a result of biophysical limitations on structure and function, a relatively small number of traits can provide key information about the functional abilities of individual species (Weiher et al. 1999). There are a number of key sources available that describe different plant traits (e.g. Ellenberg 1988; Grime, Hodgson & Hunt 1988; Fitter & Peat 1994; Hill et al. 1999; Hill, Preston & Roy 2004). However, values for each plant species are not always available for all traits, and for this reason, we use a subset of all possible traits that had a good coverage of species present within both the recreation site and the target floodplain meadows (Table 1). Following Pywell et al. (2003), the traits described in Table 1 were divided into five categories representing: (1) environmental associations (four traits); (2) life-form and history (six traits); (3) phenology (three traits); (4) regeneration (three traits) and (5) seed biology (three traits). Traits were either continuous (e.g. seed weight) or binary (e.g. either a therophyte (1) or not a therophyte (0)). These five categories have direct relevance to aspects of dispersal in both space and time (seed biology traits), establishment ability (regeneration traits) and the long-term persistence of species (environmental association, life-form and history and phenology traits), all of which represent core challenges for species survival (Weiher et al. 1999). By considering multiple traits together, it allows us to test for the relationships between functional traits, functional groups and the individual performance of species (Roberts, Clark & Wilson 2010). For each of the 19 traits, a weighted mean value (WMT) was calculated for each experimental plot in each year:
Table 1. Table of functional traits categorized for each of the species within the recreation site and the target floodplain meadow communities
Traits are divided into five main categories, and unless otherwise indicated each trait level for a species is classified as either 0 or 1 depending on whether the trait is present or absent.
Where ci is the rooted frequency of the ith plant species and Ti is the trait value of that ith species. This produced a data matrix of sites × weighted mean trait values. As weighted mean trait values differed in their size, all traits were subsequently standardized to have a mean of zero and standard deviation of 1 (Petchey & Gaston 2006). The subsequent analysis of traits focuses on these five trait categories (e.g. environmental associations) and not the individual traits from which they are composed. However, Appendix S1 in Supporting Information considers the change in individual traits over time and in response to grazing management.
Similarity to target grasslands
Recreation success was assessed for the plants, by calculating the Euclidean distance between the plant communities in each of the nine experimental plots (separately for rooted frequency and weighted mean trait values) and each of the target floodplain meadows. Note that a separate measure of Euclidean distance was calculated relative to each of the target floodplain meadow (where samples were taken from the same site in different years, these were also treated separately), and then an average Euclidean distance was calculated as a measure of recreation success. Euclidean distance was defined as:
Where: EDjk = Euclidean distance between samples j and k; Xij = either rooted frequency of species i or weighted mean trait value for trait i in sample j; WMTik = either rooted frequency of species i or weighted mean trait value for trait k; n = total number of traits. There is an inverse relationship between the Euclidean distance and the similarity of samples. This measure of Euclidean distance was also determined for all traits together (i.e. all 19 traits) and separately for the five main trait categories of environmental associations, life-form and history, phenology, regeneration and seed biology. Euclidean distance for the plants species composition was calculated only for all plants together.
For the plant species composition (as opposed to traits), Euclidean distance is a quantitative measure of similarity that would be influenced by not just the presence of a species, but also by its relative frequency (Krebs 1999). Whether or not species have simply managed to colonize a site is also of importance in terms of assessing recreation, indicating the extent to which dispersal limitation has been overcome (Bakker & Berendse 1999). For this reason, we also assessed the success of recreating plant communities using Jaccard’s binary similarity index, which considers species as present or absent only (Krebs 1999). Note that this index, which is scaled between 0 and 1, increases as similarity to the target assemblage increases. The Jaccard’s similarity index was only relevant for plant species, not functional traits.
Changes in similarity to the target grasslands between 1991 and 2009 were assessed using a general linear mixed models (PROC MIXED) for longitudinal data (Hedeker & Gibbons 2006) within SAS V9.1. Separate models were run for the response variables: (i) Euclidean distance between experimental plots and target floodplain meadows, based on the rooted frequency of all plant species; (ii) Euclidean distance based on weighted mean traits, run separately for all traits combined and individually for environmental associations, life-form and history, phenology, regeneration and seed biology and (iii) Jaccard’s similarity between experimental plots and the target floodplain meadows, based on the presence or absence of all plant species. Recreation management treatments (hay cut, cattle or sheep grazed) were included as a categorical variable, while years since the start of restoration was treated as continuous. Each maximal model included the fixed effects of treatment, year, year × year, treatment × year, treatment × year × year. The polynomial term year × year was included to identify situations where changes in similarity to the target communities were not occurring in a simple linear fashion. Random effects terms included the intercept, year and year × year (although this final random effects term was only included where the polynomial effect of year × year was retained in models as a fixed effect). To account for the repeated measures from the same plots in separate years, a categorical factor identifying individual plots was included as a subject identifier in all models, and a compound symmetry covariance structure was used. The method of parameter estimation was maximum likelihood, and model simplification was by deletion of least significant effects from the maximal model described earlier. Degrees of freedom were calculated by the iterative Satterthwaites method.
All the following analyses focus on similarity to the target grasslands at a community scale, rather than changes in the composition of individual species. A summary of general patterns of change in species composition is given in Supporting Information Appendix S2.
The first 3 years of recreation (1987–1989)
Although all the subsequent analyses were derived from samples collected between 1991 and 2009, change in community structure and trait similarity in the years preceding this (data from1987 to 1989) indicate a rapid increase in similarity during the first 3 years of recreation. This was seen for plant species composition, as measured by Jaccard’s similarity (Fig. 1), and for the similarity of all functional traits together, and separately for the regeneration and seed biology traits (Fig. 2). These data points were included in subsequent graphs as a reference point only and should be interpreted with caution as sampling methodology and experimental design pre- and post-1989 were not consistent.
Species similarity to target grasslands
From 1991 to 2009, Jaccard’s similarity index was used to measure recreation success based on the presence or absence of species, independent of their rooted frequency. When aftermath grazing management was applied, either by cattle or sheep, similarity to the target grassland increased with year (Fig. 1; Table 2). However, where aftermath grazing was absent, the slope of this change in similarity with year did not differ from zero (F1, 6·1 = 0·31, P >0·05). This pattern was explained by a significant response to both year and the interaction between year and management treatment (Table 2). There were no other significant single, interaction or polynomial terms. Where Euclidean distance was used as a continuous measure of species similarity, i.e. one that takes into account the relative frequency of individual species, there was no change in similarity to the target grassland in response to year, management or any other interaction term (Table 2).
Table 2. Results from longitudinal mixed model analysis considering temporal changes in similarity of the plant communities and functional traits to target floodplain meadow communities
Manag. × Year
Year is treated as a continuous explanatory variable. Where: β is the parameter estimate for each parameter, and βH, βC and βS are, respectively, parameter estimates for hay cutting, cattle and sheep grazing management (Manag.) where appropriate. As no significant responses to the polynomial interaction of Management × Year2 were found, this column has been excluded.
Where: NS = P > 0·05, *P < 0·05, **P < 0·01 and ***P < 0·001.
Where all 19 traits were considered together, trait similarity (Euclidean distance) to the target grasslands from 1991 to 2009 increased linearly in response to year providing grazing management (either cattle or sheep) was present in the model (Table 2; Fig 2a). Once again, where grazing was absent, the slope of the relationship did not differ from zero (Fig 2a; F1, 3·6 = 0·17, P >0·05). This pattern was explained by a significant response to year and the interaction between year and management. This same pattern was repeated for the subcategory of traits linked with plant regeneration. Similarity of the regeneration traits increased linearly with year only where cattle and sheep aftermath grazing was used, while the slope of this relationship did not differ from zero when grazing was absent (Fig 2b; F1, 2·6 = 0·07, P >0·05). No other significant effects were found for either all traits together or regeneration traits considered alone.
For the two trait subcategories of environmental associations and seed biology, similarity to the target grasslands increased at the same rate with year, independent of whether the plots were grazed or not grazed. However, the intercept for those experimental plots receiving no grazing management was higher (indicating reduced similarity to the target grasslands) than that seen for either cattle or sheep grazing (Fig. 2c,d). This was explained by a significant effect of both year and management on similarity, although no other significant interactions or polynomial terms were found (Table 2).
The similarity of traits linked with phenology to the target grasslands did not show any temporal change in response to year (Table 2). However, plots receiving aftermath cattle grazing showed greater similarity to the target grasslands, relative to the ungrazed treatment (Fig 2e). In contrast to that response seen for phenology, the similarity of traits linked with life-form and history did not show any difference in response to the effects of grazing management, either as a single term or an interaction (Table 2). However, the similarity of life-form and history traits increased from 1991 onwards, plateaued in 2002 (16 years after recreation was begun in 1986) and decreased slightly thereafter (Fig. 2f). This pattern was explained by a significant response of similarity to year and year2.
Plant species composition
Similarity to the target grasslands was seen to increase over time during recreation, although only when considering the presence or absence of colonizing species. There was a failure to replicate the relative rooted frequencies of plant species characteristic of the target floodplain meadows. This suggests that other factors may be acting to limit recreation success and community re-assembly (Bakker & Berendse 1999; Willems 2001; Pywell et al. 2002). This may include deviation from ideal abiotic environmental conditions, such as underlying hydrology and flooding regimes (Gowing et al. 2002; Matthews & Spyreas 2010), as well as differences in the initial site soil nutrient status (Janssens et al. 1998). It is also likely that priority effects relating to the timing of species establishment will have had a fundamental impact on the structure of the plant communities (Young, Chase & Huddleston 2001; Fukami et al. 2005; Young, Petersen & Clary 2005). Given that all experimental plots received the same application of seeds harvested from local floodplain meadows (McDonald 1992), such effects may be unexpected. However, the aggregated distribution of many species within swards from where seeds were harvested makes it unlikely that all species (particularly less common plants) would have been sown into all plots (Edwards et al. 2007). In addition, natural processes of colonization to the site would contribute to historical differences in the timing of species establishment, while colonizing species may not be representative of the target floodplain meadows (Young, Chase & Huddleston 2001; Bischoff 2002).
Long-term factors relating to the historical conditions under which communities developed may limit the potential for recreation to replicate what we now consider to be high-quality examples of floodplain meadows. Many of these grasslands are old, having been in existence for hundreds and even thousands of years (Burkart et al. 1998). It is possible that the environmental conditions under which they originally formed do not exist in the modern agricultural environment (Matthews & Spyreas 2010). In addition, current grazing and cutting management, while replicating perceived historic practices (McDonald 1992; Gowing et al. 2002), may bear little resemblance to management used historically during their creation. Given the probable divergence between historical conditions under which floodplain meadows were formed, it is possible that recreation would be more likely to form novel communities with at best similarities to extant ancient target grasslands (Young, Chase & Huddleston 2001; Matthews & Spyreas 2010).
Functional trait structure
While there were indications that the relative frequencies of species were not replicable, recreation management was seen to have greater success in reproducing the functional trait structure of the target grasslands (Wilson 1999; Temperton et al. 2004; Fukami et al. 2005). Traditional forms of grassland management (i.e. late hay cutting followed by aftermath livestock grazing by cattle or sheep) will dictate available niche space, which will select for specific functional traits from the pool of colonizing and already established species (Young, Chase & Huddleston 2001; Fukami et al. 2005). We find supporting evidence for such a process, as shifts over time in the similarity of functional traits towards those of the target floodplain meadows were typically linked to the presence of grazing management. This pattern of increasing similarity over time was seen for four of the five subcategories of traits considered, with only traits linked with plant phenology not increasing in similarity over time. It should also be noted that traits linked with life-form and history showed some evidence that they were diverging from those of the target meadows during the last 5 years of the study. It is possible that this may indicate that similarity has plateaued for these traits and is not actually decreasing in similarity to those seen for the target floodplain meadows. However, either trend would indicate that replication of the target community, at least in terms of life-form and history traits, may be unlikely to occur. When all functional traits are considered together, then, under typical grazing management, it would take 74 (SE ± 9) years to replicate the functional structure of the plant traits assuming that the linearity of the observed trends continues. In contrast, there is no evidence that replication of species composition in terms of relative rooted frequencies will ever occur. This is in agreement with chrono-sequences that suggest that after 60 years, limited similarity between recreated and ancient grassland communities in terms of species composition is achieved (Fagan et al. 2008).
The potential for re-establishing functionally equivalent grasslands, even over this time-scale, has important implications when considering their capacity to deliver ecosystem services (Benayas et al. 2009). Such services, while not considered in the current study, may include the production of hay and meat, reduction of soil erosion, carbon sequestration and enhanced infiltration of water that may help to mitigate against flood risk elsewhere in the floodplain (Posthumus et al. 2010). The extent to which recreated sites need to replicate the functional structure of ancient floodplain meadows to deliver these ecosystem services is likely to be low. Independent of this, even small shifts in similarity of functional traits towards those of the target communities may have important consequences for the stability of these grasslands (Tilman 1997). For example, colonization by pernicious weed species (i.e. those considered to have deleterious economic consequences for agriculture) may be reduced in grasslands that show an increase in functional diversity comparable to that of the target habitats (Vitousek & Hooper 1997; Westbury et al. 2008; Young et al. 2009). Conversely, the rapid establishment of undesirable species early during recreation may well have the reverse effect, by reducing the chance that functionally similar plants to those of floodplain meadows subsequently established (Vitousek & Hooper 1997; Young et al. 2009).
Impacts of management
Management in the form of aftermath grazing was shown to be a key element promoting recreation, not just from the perspective of species colonization, but also in terms of functional trait similarity to the target grassland. Indeed, for species colonization, overall traits and the regeneration traits subcategory, an increase in similarity to the target floodplain meadows only occurred where aftermath grazing was used (with no change in similarity where grazing was absent). Although a simplified interpretation of the complex effects of grazing, this shift in functional similarity could have resulted in part from selective foliage removal as a result of livestock feeding (Bullock et al. 2001; Rodriguez et al. 2003; Rook et al. 2004). For example, where grazing is prevented in grasslands, there is typically a shift in the functional structure of plants towards rhizomatous species, erect species and species >30 cm in height (Rodriguez et al. 2003). Given that aftermath grazing represents historical management for the target floodplain meadows, its importance as a management tool during recreation was not unexpected (Gowing et al. 2002). Grazing livestock may also play an important role in the dispersal of seeds (Fischer, Poschlod & Beinlich 1996) and as such may help to establish plants typical of target grassland communities. This is particularly so if livestock are moved between different fields, especially if one field is floristically species rich. However, prior to grazing of the experimental plots in this study, cattle and sheep would have been grazed on species-poor improved grasslands. Livestock also remained in place on individual plots while grazing, and as such their capacity as a dispersal agent for seeds is likely to have been low.
After 22 years of management, complete recreation of a floodplain meadow still remains an elusive goal, both in terms of plant community structure and the re-establishment of functional trait characteristics. This highlights both the practical limitations that must be overcome during recreation (Bakker & Berendse 1999; Willems 2001; Pywell et al. 2003; Woodcock et al. 2010), as well as emphasizing the need to protect high-quality meadows that have escaped degradation or destruction. While recreation can be argued to have increased the biodiversity value of what was previously arable land, its failure to establish the desired communities places into a negative light any government policies that emphasize mitigation and compensation measures following planned developments on existing meadows (Matthews & Spyreas 2010). Specifically, if the linear trends presented here continue, then it would take c. 162 years (SE ± 26 years) to fully restore floodplain meadows, and this would only be in terms of getting appropriate species to establish. It should be noted that this time-scale assumes restoration of species composition relative to the idealized target community used in this study. Given dissimilarity in species composition between floodplain grasslands, restoration representing the establishment of 75% of species in this target community may well represent a more realistic goal. Even so, this would still take c. 94 years (SE ± 14 years). Independent of this caveat, if this is the time-scale needed to achieve recreation, then any compensation scheme proclaiming they can replace floodplain meadows lost to development (i.e. gravel extraction) is being wholly unrealistic. Ultimately, recreation should serve as a tool to augment and buffer existing areas, while as a compensatory measure, its value should be carefully considered in the light of other options.
Thanks to David Roy for help with plant traits and to the University of Oxford’s Department of Plant Sciences, Oxford University Field Station and farm staff, particularly Anna Winton and Ruth Layton of the Farm Animal Initiatives Ltd. Permission to sample at Oxey and Yarnton Meads was kindly given by Paul Allen (BBOWT).