Do restored calcareous grasslands on former arable fields resemble ancient targets? The effect of time, methods and environment on outcomes

Authors


*Correspondence author. E-mail: katecf@hotmail.com

Summary

  • 1A great deal of money is being invested in calcareous grassland restoration on arable land within agri-environment schemes in the European Union. There is, however, little evidence that the target ecosystem can be obtained from the restoration techniques and management practices currently used. We evaluated these techniques using a multi-site approach in order to improve the success of future restoration efforts.
  • 2We compared 40 restoration sites with 40 paired reference sites and addressed the following specific hypotheses: (i) Are plant communities of restoration sites becoming more like those of mature calcareous grassland? (ii) How long does the restoration process take? (iii) Are there any environmental filters that hinder the process? (iv) Is there a difference in plant attributes between restored and ancient grassland communities, and between restored communities of different ages?
  • 3We used a multivariate approach to assess the similarity of sites and found that there was little overlap between restored and ancient grassland communities even after 60 years. Successful restoration of calcareous grasslands is achievable but the process is slow.
  • 4Different plant attributes were present at different frequencies in restored and reference sites, and the frequency of some attributes became more like those of reference sites with increasing age of restored site (e.g. perenniality and ruderality).
  • 5Seeding restoration sites with a low diversity mix appeared detrimental to restoration. Sites that regenerated naturally moved towards the target over time, although success was limited by proximity to ancient grassland vegetation. High soil phosphorus concentration was detrimental to restoration.
  • 6Synthesis and applications. We recommend selecting restoration sites with low phosphorous concentrations that adjoin patches of ancient calcareous grassland. Seed mixes should be devised carefully to prevent the assembly of low-value, competitive, stable communities dominated by grasses; natural regeneration may avoid this, but will only be effective close to sources of propagules. Other methods of restoration or habitat management would undoubtedly benefit from similar multi-site evaluation.

Introduction

Calcareous grassland is said to be the most diverse habitat in Europe (Poschlod & WallisDeVries 2002), supporting many rare plants and invertebrate species. It is of high conservation value and has great aesthetic appeal. Calcareous grasslands once covered large areas of Western Europe but these have diminished during the 20th century through intensification of agriculture and cessation of traditional management practices (Keymer & Leach 1990; Hutchings & Booth 1996; Critchley, Burke & Stevens 2003). In Britain, it is estimated that 80% of unimproved chalk grassland present in 1939 has been lost (Keymer & Leach 1990), with a similar proportion lost from limestone (Newbold 1989).

One of the key aims of European Union conservation policy, as reflected in its Habitats Directive (European Union 1992), is the maintenance and enhancement of calcareous grasslands. In line with this, a budget of over €2 billion was allocated for use in agri-environment schemes by the European Union in 2006. The UK Biodiversity Action Plan (Anonymous 1995) has set a target of re-establishing 1000 ha of lowland calcareous grassland of wildlife value by 2010. In order to achieve this, agri-environment schemes provide grant aid for the management of unimproved calcareous grassland and the restoration of this habitat on ex-arable land.

There is conflicting information on the success of restoration schemes on calcareous soils. There is evidence that grassland communities can reach equilibrium on land that has been ploughed, although it takes a relatively long time (Wells et al. 1976). Others have queried the value of restoration as a conservation strategy in such situations (Dobson, Bradshaw & Baker 1997), since the process is lengthy and there are no reports of definite success.

The way in which an ecosystem is assembled is likely to influence how much it resembles the target system. Most restoration ecologists regard the alternative stable states (ASS) model of community development to be most accurate (Bradshaw 1983; Suding, Gross & Houseman 2004; Didham, Watts & Norton 2005); that is, there are a number of possible species assemblages and certain biotic or abiotic factors (or their temporal order of occurrence) can cause a switch between them. In practical terms, these factors can be interpreted as ‘constraints’ or ‘filters’ to restoration success. Possible constraints are high soil fertility, propagule limitation, absence of keystone species and inappropriate mowing or grazing regimes (Bakker & Berendse 1999; Walker et al. 2004).

In order to make useful evaluations of restoration success, it is imperative for the end point to be identified (Zedler & Callaway 2000; McCoy & Mushinsky 2002; Wolters, Garbutt & Bakker 2005). In most cases, local reference sites will provide much of the necessary information (White & Walker 1997; Ruiz-Jaen & Aide 2005); here, we use local undisturbed and unimproved calcareous grasslands as examples of target ecosystems. This helps to reduce sources of error inherent in the chronosequence approach by controlling for local variation in factors such as climate and species pool. A chronosequence is necessary for this study due to the length of time necessary for semi-natural grassland restoration and the lack monitoring and experiments on this time-scale (Bakker et al. 1996).

In many cases, evaluations of restoration success are based on plant communities alone (Walters 2000; Wilkins, Keith & Adam 2003; Voigt & Perner 2004). While this is a simplification, feedback mechanisms within the species interactions with abiotic conditions and with other trophic levels should provide useful summaries of the communities as a whole, and an indication of how they are functioning (Bakker et al. 2000). However, the greater the number of ecosystem measurements made, the greater the chance of information leading to an improved restoration method.

Here, using a novel multisite approach spanning a time-scale of 60 years and considering selected environmental conditions, we attempt to establish whether calcareous grassland plant communities under restoration management are moving towards those of target ecosystems. Various factors might be expected to impinge on restoration success, particularly: (i) elapsed time since restoration, where we hypothesize that success increases with time (Zedler & Callaway 1999; Willems 2001) due to succession and increasingly beneficial soil conditions from nutrient loss and positive feedbacks; (ii) seeding of restored sites, where we hypothesize that seeding with a diverse seed mixture produces a system more like the target system more quickly than naturally regenerating sites (Wells 1990; Pywell et al. 2002; Walker et al. 2004); (iii) distance to nearest undisturbed calcareous grasslands as potential donor sites for propagule dispersal, where we hypothesize that success increases with shorter distances (Gibson & Brown 1991); (iv) abiotic environmental variables, particularly soil fertility, where we hypothesize that the least fertile sites will be more successful (Gough & Marrs 1990; Marrs 1993) since a characteristic of species-rich calcareous grassland is its low fertility (Willems & van Nieuwstadt 1996). We also test a fifth hypothesis, that plant attributes differ in their abundance between ancient and restored communities; from subsequent comparisons, we aim to provide explanations for the community structure development and further indications of how it is functioning (Hodgson 1990; Lavorel & Garnier 2002; Pywell et al. 2003; Kahmen & Poschlod 2004).

Methods

site selection

A data base was compiled of all calcareous grassland ex-arable sites where restoration had been attempted in five regions in southern England (North Downs, South Downs, South Wessex Downs, Chilterns, Cotswolds); sites were grouped on the basis of age of restoration or time since arable agriculture ceased (i.e. 1–5 years, 6–10 years, 11–20 years and > 20 years). From the pool of available sites, 40 restoration study sites were selected randomly so that there were approximately equal numbers of sites in each region and age class. The closest ancient grassland (defined as having been under permanent grassland for at least 200 years) with a similar slope angle and aspect to the selected restoration site was selected as a paired reference (Supplementary Material Figure S1). All sites, restoration and reference, were grazed, although occasionally this was supplemented by mowing. In many sites, disturbance management was determined by the associated agri-environment schemes.

At each site, the following information was recorded: (i) coordinates (eastings and northings), aspect, altitude, and approximate angle of the slope, area of restoration, and the distance between the restoration and reference site; and (ii) information on restoration methods, including a list of sown species, collected from the landowners and managers of each site. The area of restoration sites (between 1 ha and 103·4 ha) and distance between paired sites (from adjacent to 9·3 km) were calculated using the Defra electronic map (http://www.magic.gov.uk) applying the ‘measure area’ and ‘measure distance’ functions, respectively. The information on restoration methods was used to classify the sites into three seeding classes: natural regeneration (no seed added), seeded with grasses only (this category also includes those sites with either Lotus corniculatus or Trifolium repens added to the grass mix), and seeded with a more diverse mix of grasses and forbs. Generally, diversely seeded sites were sown with seed harvested from nearby areas of ancient calcareous grassland, sometimes as green hay. A few sites used specially formulated seed mixes, with the most diverse consisting of 36 forb and 10 grass species and the least diverse five forb and seven (commercial) grasses.

survey methods

At each pair of sites (reference and restored), a 100-m transect was located with, as far as possible, the same aspect and slope angle, and at least 10 m from field boundaries, field size permitting. The use of a transect enabled surveys to be made at equal scales at all sites while providing a reasonable measure of within-site variation. Vegetation surveys were carried out between June and August 2004, with paired reference and restoration sites surveyed on the same day or on consecutive days. These were carried out using a 50 × 50 cm quadrat at 10 points along the transect, 10 m apart. At each sampling point the percentage cover of each vascular plant species was estimated as a vertical projection. Nomenclature of vascular plants follows Stace (1997).

In early September 2004, 15 soil samples were taken at equal distances along the same transect using an auger to a depth of 15 cm; soil was taken from the equivalent reference site on the same day as from the restoration site. All soil samples taken from any one site were combined, then analysed for pH, total carbon, total nitrogen (Dumas combustion method), exchangeable ammonium nitrate extractable sodium, calcium, magnesium and potassium, and Olsen's phosphorus (Allen 1989).

statistical analysis

Analysis of plant community data was performed with canoco for Windows version 4.5 (ter Braak & Šmilauer 2002). All analyses were performed using transformed [log(y + 1)] species data. Twenty-two species found only once were excluded from the analysis. Initially, Detrended Correspondence Analysis (DCA) was used to measure gradient lengths to assess whether a unimodal or bimodal model should be used. All ordinations used the mean species cover per site, calculated from quadrat data, and mean values per site for each environmental variable (apart from ordinations leading to the ellipses in Supplementary Material Figure S2 where individual quadrat data were used).

This DCA produced an ordination with a first axis gradient length of 3·5 SD units, justifying the use of the unimodal model (i.e. the data were too heterogeneous for a linear response) (Lepš & Šmilauer 2003). The distribution of different sites within the species ordination produced was examined using the scores produced for axes 1 and 2 and calculating from these the Euclidian distance between each restoration site and its associated reference site.

Canonical Correspondence Analysis (CCA) was used to initially explore relationships between species and environment data. The Forward Selection procedure was used with a Monte Carlo test with 499 permutations to assess which environmental variables were significant. Reference/restoration and seeding categories did not affect the analysis, but were included in the figure for illustration purposes. Further exploration of the relationship between the Euclidian distance and age of the restoration site [both ln(y + 1)-transformed], soil N, P, K, C, Na, and Mg concentrations, soil pH, altitude, latitude, longitude, and area and isolation of restoration site (using Johnson transformation if necessary, Minitab version 14) was carried out using regression analysis in Minitab (Ryan, Joiner & Ryan 2000). Multiple regressions were carried out for the combined data and the data split into seeding categories, with models selected by the best subsets method from all environmental variables investigated in the simple regression. ancova (Minitab GLM) was used to examine any significant differences in the regression lines for sites seeded in different ways.

For illustration, reference site data were summarized with a bivariate standard deviational ellipse (SDE) using the Excel algorithm detailed by Milligan et al. (2004), from one produced by Ricklefs & Nealen (1998); this provides an estimate of the relative diversification within or between sites based on the first two axes produced by ordination. Using the same method, SDEs were produced for individual reference and restoration sites to compare those restored sites seeded differently and of different ages (Supplementary Material Figure S2).

Further to the multivariate analysis, several univariate investigations were made in Minitab. Correlation analysis was carried out between all the environmental variables mentioned above. In addition, we tested for differences in the occurrence of individual species between the reference and the restoration sites using t-tests from the species cover data (arcsin(√y/100)-transformed). Because of the large number of calculations, we accept that a few of these would be expected to be significant by chance alone; thus, we accept that some significant differences detected may be unsafe (this also applies to attribute calculations in the next section).

analysis of plant attributes

The literature was searched for comprehensive basic plant attribute data (Table 2). In addition, we used Grime's CSR classification of plant strategies (Grime 1979) and Ellenberg's classification of species according to their nutrient requirements, modified for Britain (Hill, Preston & Roy 2004). Before use, seed bank persistence data were transformed into an index following Thompson et al. (1998), with species with fewer than five records excluded from the index. These data are an indication of the likelihood that the species has persistent seed rather than a measure of length of persistence of its seed bank.

Table 2.  Plant attributes differing significantly between reference sites and restoration sites. Change index indicates the relative magnitude of change seen between the period 1930–1960 and 1987–1999; GB range indicates the number of 10 km squares in Britain in which a species is found; Broad habitats indicates the number of main (as opposed to minor) habitats of a species. Dispersule weight and shape were tested but were found to be insignificant in their difference between reference and restoration sites. For attributes Life form, Regeneration strategy and Dispersal/germinule form, other categories were tested and found to be insignificantly different. NS, not significant (P > 0·05); G, grass-seeded sites; D, diversely seeded sites; NR, naturally regenerated sites
 SourceMean frequency in reference sitesMean frequency in restored sitesTest statisticPCorrelation with environmental variablesEffect of seeding category (all means show the reference site value minus the restored site value)
anova statisticsG meanD meanNR mean
  1. *Data sources: 1. Hill et al. (2004); 2. Hodgson et al. (1995); 3. Thompson et al.

Change index10·0040·230T = –7·000< 0·000Phosphorus: r = –0·354, P = 0·025F = 5·93, P = 0·006–0·398–0·182–0·155
GB range11473·131739·34T = –4·67< 0·000 F = 5·30, P = 0·010–432·9–366·3–74·4
Number of habitats11·0431·200T = –4·66< 0·000 NS   
Ellenberg N12·5553·323T = –7·15< 0·000 F = 3·30, P = 0·048–1·193–0·811–0·465
% rare10·0040·000T = 2·910·006 NS   
% perennial10·9180·802T = 6·69< 0·000Site age: r = –0·529, P < 0·000NS   
      Nitrogen: r = –0·415, P = 0·008    
% woody10·0520·014T = 4·80< 0·000 NS   
Competitiveness20·1660·222T = –6·04< 0·000 NS   
Stress tolerance20·4040·219T = 7·68< 0·000Distance: r = 0·363, P = 0·021NS   
Ruderality20·1640·302T = –9·96< 0·000Site age: r = 0·459, P = 0·003F = 8·82, P = 0·001–0·213–0·141–0·088
      Phosphorus: r = –0·387, P = 0·014    
      Nitrogen: r = 0·449, P = 0·004    
Seed bank longevity30·2540·327T = –6·97< 0·000Site age: r = 0·320, P = 0·044NS–0·098–0·093–0·041
      Phosphorus: r = –0·368, P = 0·020    
      Nitrogen: r = 0·384, P = 0·014    
Life form: hemicryptophytes20·8570·770T = 4·93< 0·000 NS   
Life form: therophytes20·0000·149T = –11·76< 0·000Site age: r = 0·476, P = 0·002NS   
      Phosphorus: r = –0·378, P = 0·016    
Canopy height21·5551·656T = –2·120·041 NS   
Regeneration strategy: seed bank20·2490·322T = –4·57< 0·000Site age: r = 0·396, P = 0·011NS   
      Nitrogen: r = 0·335, P = 0·035    
Regeneration strategy: vegetative20·3600·209T = 6·69< 0·000Site age: r = − 0·384, P = 0·015NS0·2120·1850·084
      Nitrogen: r = –0·406, P = 0·009    
Regeneration strategy: numerous seeds20·1390·077T = –4·25< 0·000Distance: r = –0·355, P = 0·024    
 NS    Site age: r = 0·358, P = 0·023    
Agency of dispersal: animal20·3470·316T = 2·320·026Distance: r = 0·329, P = 0·038NS   
      Phosphorus: r = 0·395, P = 0·012    
      Latitude: r = 0·410, P = 0·009    
Agency of dispersal: wind20·1510·211T = –3·260·002Site age: r = 0·334, P = 0·035NS   
      Site area: r = –0·396, P = 0·011    
Dispersal/germinule form: fruit20·7350·669T = 4·17< 0·000Site age: r = –0·326, P = 0·040NS   
Dispersal/germinule form: seed20·2130·169T = 3·270·002 F = 4·76, P = 0·0150·1090·0140·029
Dispersal/germinule form: spore20·0020·000T = 2·420·020 NS   
Dispersal/germinule form: dry seed/seedpod20·0500·162T = –8·14< 0·000 NS   

For each species at each site, the plant attributes represented by continuous values were multiplied by the frequency (i.e. proportion of quadrats in which present) of that species at that site. From this, a mean value for each plant attribute at each site was derived. For categorical data, the proportion of each category was found for each site, and again this was multiplied by the respective frequency. Paired t-tests were used in both cases to find differences in attribute frequency between restored sites and reference sites. Some data were unavailable; thus, not all species were used in the analysis of every attribute.

The data for each plant attribute at each restored site were subtracted from the data for the corresponding reference site to find the difference. Correlation coefficients were then used to find any relationship between plant attributes and age of restoration, as well as other environmental characteristics found to be important in previous analyses. The effect of different seeding strategies on plant attributes was measured using ancova, excluding any age effect.

Results

Unconstrained ordination of sites showed an overall pattern of restored sites becoming closer to reference sites with increasing restoration site age (Euclidian distance = 1·49 –(0·3 × restoration age), F = 5·78, P = 0·021, explaining 13·2% of the variation in Euclidian distance), although there is very little overlap even after 60 years (Fig. 2 and Figure S2, Supplementary Material).

Figure 2.

The distribution of sites on the first two axes of the DCA: (a) compares the ancient calcareous grassland sites inline image and the restored sites inline image; (b) compares restored sites of different age classes: inline image = 0–5 yrs; inline image = 6–10 yrs; inline image = 11–20 yrs; inline image = 20–60 yrs; (c) compares restored sites of different seeding categories: inline image, sites restored with grass species and a maximum of one forb; inline image, sites seeded with grass and forb species; inline image, sites that have regenerated naturally. In (b) and (c), reference sites are represented by SD-ellipses.

The species distribution (Fig. 1) and species abundance data (Table 1) showed that species typical of chalk grassland [e.g. Helianthemum nummularium, Viola hirta, Pilosella officinarum, Thymus polytrichus and Scabiosa columbaria, all of which are used by the UK's Department of Environment, Food and Rural Affairs (Defra 2005) as calcareous grassland indicator species] were clustered around the reference sites’ centroid, and generalist species found by Gibson & Brown (1991) to be typical of early successional grassland (e.g. Lolium perenne, Medicago lupulina, Trifolium repens, Taraxacum officinale agg. and Ranunculus repens) around the restored sites’ centroid. A much broader spread in species was found in restoration sites (Fig. 2), which can be at least partly explained by the seeding strategy (Fig. 2c). Sites seeded with grasses (and one or two commercial legumes) are located in the top left of the graph; these sites were typically sown with seed mixtures containing Lolium perenne, Phleum bertelonii, Agrostis capillaris and Cynosurus cristatus, all species located in the corresponding area of Fig. 1. Similarly, Trifolium repens can be seen in the same area. These grass-seeded sites remained farthest from the reference sites. Sites which were seeded with a diverse seed mixture and those that had regenerated naturally cannot be differentiated. However, some young restoration sites sown with a particularly diverse seed mix were much closer to the reference sites than the overall trend, for example, the most diverse specially formulated mix mentioned in the methods.

Figure 1.

CCA biplot showing species and environmental variables calculated as significant (P < 0·05) by Forward Selection. All species are referred to by the first four letters of the genus name followed by the first four letters of the species name: Achimill, Achillea millefolium; Agrostol, Agrostis stolonifera; Aspecyna, Asperula cynanchica; Arrhelat, Arrhenatherum elatius; Bracpinn, Brachypodium pinnatum; Brizmedi, Briza media; Bromerec, Bromopsis erecta; Camprotu, Campanula rotundifolia; Carecary, Carex caryophyllea; Careflac, Carex flacca; Cerafont, Cerastium fontanum; Cirsacau, Cirsium acaule; Cirsvulg, Cirsium vulgare; Cratmono, Crataegus monogyna; Crepcapi, Crepis capillaris; Crepvesi, Crepis vesicaria; Cynocris, Cynosurus cristatus; Dactfuch, Dactylorhiza fuchsii; Dactglom, Dactylis glomerata; Euphnemo, Euphrasia nemorosa; Festprat, Festuca pratensis; Filivulg, Filipendula vulgaris; Helinumm, Helianthemum nummularium; Heliprat, Helictotrichon pratense; Helipube, Helictotrichon pubescens; Holclana, Holcus lanatus; Koelmacr, Koeleria macrantha; Leonautu, Leontodon autumnalis; Lolipere, Lolium perenne; Medilupu, Medicago lupulina; Odonvern, Odontites vernus; Pastsati, Pastinaca sativa; Pilooffi, Pilosella officinarum; Pimpsaxi, Pimpinella saxifraga; Planlanc, Plantago lanceolata; Ranubulb, Ranunculus bulbosus; Ranurepe, Ranunculus repens; Rosaarve, Rosa arvensis; Sangmino, Sanguisorba minor; Scabcolu, Scabiosa columbaria; Soncaspe, Sonchus asper; Taraoffi, Taraxacum officinale agg.; Thympoly, Thymus polytrichus; Trifprat, Trifolium pratense; Trifrepe, Trifolium repens; Veroserp, Veronica serpyllifolia; Violhirt, Viola hirta. Squares indicate those species used by Defra (2005) as calcareous grassland indicator species, circles grass and sedge species, and triangles the remainder. inline image represents the centroid of reference sites, inline image of restoration sites, NR of naturally regenerated sites, G of grass-seeded sites and D of diversely seeded sites. The first four eigenvalues of the ordination were 0·291, 0·119, 0·086 and 0·071.

Table 1.  Species found to have a significantly different abundance between reference and restoration sites as found using paired t-tests. *, significant at the 5% level; **, significant at the 1% level; df = 39 for each species
Species more abundant in reference sitesSpecies more abundant in restored sites
Speciest-valueSpeciest-value
Achillea millefolium2·56*Agrostis stolonifera4·19**
Anthoxanthum odoratum2·65*Bromus hordeaceus2·33*
Asperula cynanchica4·11**Cerastium fontanum2·90**
Brachypodium pinnatum3·91**Crepis capillaris2·84**
Briza media3·66**Cynosurus cristatus4·42**
Bromopsis erecta4·30**Daucus carota2·61*
Campanula rotundifolia4·37**Geranium columbinum2·56*
Carex caryophyllea4·34**Geranium molle2·73**
Carex flacca11·86**Lolium perenne5·24**
Centaurea nigra4·33**Medicago lupulina3·84**
Cirsium acaule4·92**Ranunculus repens2·74**
Crataegus monogyna2·57*Taraxacum officinale agg.4·33**
Euphrasia nemorosa2·10*Trifolium repens4·88**
Festuca rubra/ovina2·97**Veronica serpyllifolia2·05*
Filipendula vulgaris3·70**  
Galium verum3·44**  
Helianthemum nummularium4·06**  
Helictotrichon pratensis5·47**  
Helictotrichon pubescens5·92**  
Koeleria macrantha5·7**  
Linum catharticum3·69**  
Lotus corniculatus3·92**  
Pimpinella saxifraga4·48**  
Pilosella officinarum4·10**  
Polygola vulgaris2·24*  
Ranunculus bulbosus2·53*  
Sanguisorba minor8·90**  
Scabiosa columbaria3·98**  
Succisa pratensis2·17*  
Thymus polytrichus4·50**  
Viola hirta5·17**  
Total31 14

ancova showed that the effect of seeding category was still significant after eliminating the effect of age. Only the Euclidian distances from one of the seeding classes were significantly associated with age: the naturally regenerated sites [Euclidian distance = 1·50 – (0·521 × restoration age), F = 17·0, P = 0·001, R2 = 54·8%]. Although slopes of the three regression lines were not significantly different from one another (P = 0·172), the intercepts were (F = 6·68, P = 0·003, R2 = 36·69%). However, ancova using sites from just the naturally regenerated and the diversely seeded categories showed that there was no difference between the two when the effect of age was eliminated (P = 0·333).

There was no significant relationship between the physical distance between the restoration site and the nearest area of ancient calcareous grassland and the Euclidian distance (F = 3·43, P = 0·072), but a significant negative relationship was found for both the naturally regenerated (Euclidian distance = 0·873 + (0·000903 × actual distance), F = 7·82, P = 0·014, R2 = 35·9%) and grass-seeded sites (Euclidian distance = 1·84 + (0·000214 × actual distance), F = 10·48, P = 0·009, R2 = 51·2%) when analysed separately. No relationship was found for sites seeded with both grass and forb species. ancova showed that both the slopes and the intercepts of these three lines were significantly different (F = 7·03, P = 0·003 and F = 6·56, P = 0·004, respectively).

Simple regression indicated that the difference in P concentration is able to explain the greatest variation in Euclidian distance between site pairs (F = 10·57, P = 0·002, R2 = 21·89%). P concentration was also a prominent factor in the CCA (Fig. 1). Latitude was the only other environmental variable to show a significant effect on restoration success in simple regression of the total data (F = 7·74, P = 0·008, R2 = 14·7%). Other soil variables were significant in their explanation of site ordination (sodium, magnesium, potassium and pH) (Fig. 1), despite inter-correlation (Supplementary Material Table S3). Many soil chemical concentrations were also correlated with age, suggesting that soil conditions of restoration sites became more like those of reference sites over time.

Multiple regression provided an overview of the effects of environmental variables. The data combined from all sites produced the following model:

image

When the different seeding categories were analysed individually, the following results were obtained:

image
image
image

All of these results explained a much greater proportion of variation than the model for the total data. Particularly evident in both the naturally regenerated and diversely seeded sites is the effect of proximity of the restoration site to established calcareous grassland.

Plant community attributes also differed between sites (Table 2). As might be expected due to high residual fertility in recently abandoned arable land, restoration sites had a higher mean Ellenberg N value than reference sites. Reference sites had a greater frequency of rare plants with smaller geographic ranges that were on average shorter in height and more likely to be woody than restoration-site communities. Overall, plants found in reference sites were more perennial and hemicryptophytic, and were more likely to be dispersed as a fruit by animals or rely on vegetative regeneration. The most common regeneration strategy of restored site plants was by seed bank; their seed banks were more persistent than those of ancient calcareous grassland. The seeds had a greater likelihood of being transported by wind, and plants were less likely to produce fruits and more likely to have a therophytic life form. The most important strategy in ancient grasslands was stress tolerance, while in restored sites, there was a greater degree of competitiveness and ruderality.

There is some evidence that plant attributes in restoration sites become more like those of ancient calcareous grassland sites over time. Species in restored sites showed a tendency to become less ruderal and more perennial over time. Their seed banks became less persistent, they became less reliant on their seed banks in regeneration, their seeds became less likely to be wind-dispersed and they became more likely to produce fruits. Several attributes showed corresponding correlation with soil chemical concentrations, as expected due to the correlation between these and restoration site age. Some attributes were correlated with the distance between restoration sites and the nearest area of ancient calcareous grassland; a regeneration strategy of numerous seeds was more common in sites farther from ancient grassland, while dispersal by animals was more common in sites closer to ancient grassland, as was stress tolerance.

Discussion

does restoration success increase with time?

On the whole, plant communities of older restoration sites are more like those of their reference sites than younger restoration sites, although in general, the process exceeds the 60 years covered by this study (Supplementary Material Figure S2). There is evidence that ex-arable land on calcareous soil will succeed to species-rich calcareous grassland if the circumstances are right.

what effect does seeding strategy have on restoration success?

Diversely seeded sites are not significantly closer to their associated target sites than naturally regenerated sites, while grass-seeded sites are significantly less successful. When analysed individually, only naturally regenerated sites became more like ancient calcareous grassland sites over the time-scale investigated, suggesting that seeding may suppress succession. No seed mix will ever contain the exact community that would result from succession, which under the assembly rules theory is a potential problem (Lockwood & Samuels 2004). If, as commonly agreed, the ASS model bears the closest resemblance to reality, seeding with ecologically inappropriate species (or proportions of species) could cause the trajectory to veer in the wrong direction. An ASS would result, as demonstrated in Fig. 2c by the cluster of grass-seeded plots on which time since restoration appears to have no effect.

We hypothesized that seeding with a diverse seed-mix would be beneficial to restoration, but there is no evidence to show that the limited propagule constraint has been overcome by seeding. This conflicts with the results of the more experimental approach taken by Pywell et al. (2002) and Kiehl, Thormann & Pfadenhauer (2006), and of smaller-scale studies such as Edwards et al. (2007). It is possible that the benefits of the application of carefully considered diverse seed mixes have been masked by inappropriate mixes that fell into the ‘diversely seeded’ category. In addition, disturbance and sowing techniques surrounding seed application have been found to make a difference to plant establishment (Edwards et al. 2007), an aspect of restoration which our study did not investigate.

The chronosequence approach is particularly useful in studying long-term processes in ecology (Foster & Tilman 2000). It can, however, be prone to errors. We were able to minimize the local variation by using a local reference site for each restoration site, but the possibility that site factors could be covariable with age (Bakker et al. 1996) is much more difficult to mitigate. The restoration sites dating from the 1940s and 1950s would have been under much less intensive arable cultivation than post-1970s sites, of considerable importance considering the effects of soil nutrient concentrations. It is likely that older restoration sites had starting points more conducive to successful restoration than younger sites, and naturally regenerating sites are the oldest of the sites studied here (Supplementary Material Figure S2). Seeding to accelerate restoration has been carried out only relatively recently.

does proximity to sources of colonists alter restoration success?

Naturally regenerated sites showed a dramatic positive response to the closeness of good quality grassland. Due to the very short-lived nature of calcareous grassland seed banks, these sites must be within potential dispersal distance of calcareous grassland species for restoration to achieve any degree of success. Since calcareous grassland species are likely to propagate vegetatively, they may well have extremely small dispersal distances. Responses by those sites seeded just with grasses were not so pronounced but were actually even more significant, so were apparently still able to support the growth of newly germinated seedlings arriving naturally. It is possible that the lack of any positive relationship when sites are seeded with both grasses and forbs is due to a lack of gaps for colonization (multiple regression showed an apparently adverse effect of proximity to sites of ancient calcareous grassland, a result so unlikely that we consider it a Type II error). However, since seeded sites are designed to bypass the early successional stages, propagules from the seed rain are much less important, and the relationship between proximity to ancient calcareous grassland and restoration success should be weaker and more difficult to detect.

how do other environmental variables affect restoration success?

Restoration success was very dependent on environmental variables, but these were so inter-correlated that the causal relationships are difficult to establish. Nevertheless, it is clear that although a large proportion of the variation of diversely seeded and naturally regenerated sites can be explained in ways that suggest restoration will proceed over time, the same is not true of grass-seeded sites.

Time since restoration exhibited a significant independent effect, but this seems at least partly due to P concentration, in agreement with evidence found by Critchley et al. (2002a), Janssens et al. (1998) and Prach (2003). N (total rather than available) was generally lower in restoration sites than reference sites, due to the low organic matter present in restored sites. Calcareous grassland restoration sites suffer from a lack of available N once initially high levels have leached from the soil (Gough & Marrs 1990). It is clear that if soil nutrient concentrations are a factor in restoration success, a high phosphorus level is the constraining factor rather than general nutrient enrichment.

Calcareous grasslands are generally considered to be P-limited systems (Johnson, Leake & Lee 1999; Critchley et al. 2002b), containing many plant species adapted to low P conditions, for example, by associations with mycorrhiza, cluster roots and phosphatase exudation (Adams et al. 2004; Wassen et al. 2005). Increased P due to fertilization persists for many years, resulting in an N-limited system encouraging different plant species (Roem & Berendse 2000) and selecting graminoids and stress-tolerators (Güsewell 2004). While the system is N-limited rather than P-limited, the plant communities will reflect this. Although P levels were variable between reference sites, it is not the amount of P but whether the system is N- or P-limited (i.e. the N:P ratio) that is important. Only a few of the restoration sites used here had similar P levels to those of their respective reference sites, suggesting that on average they will need around 50 years or more to return to an acceptable level. This must be dependent on the level of fertilization while they were arable, however.

is there any effect of restoration on plant community attributes?

Typically, plant attributes found more commonly in ancient calcareous grassland indicate a degree of specialization (animal-dispersed fruits), longevity and stress-tolerance (woody plants and hemicryptophytes). In contrast, ex-arable soil favoured opportunistic, generalist species that thrive on soil where competition for nutrients and space is not yet established (Owen & Marrs 2000), that is, short-lived species of high reproductive rate and few specific habitat requirements. Attributes that epitomize this are wind dispersal of seeds and reliance on a seed bank for reproduction. In general, species found in reference sites were typical of climax or late-successional communities. The frequency of so many attributes in restoration sites becoming more like that found in reference sites with age indicates that succession is proceeding in the direction of the target ecosystem. There are also attributes that indicate the effect of distance between the restoration site and the nearest area of good quality grassland: stress-tolerators with animal-dispersed seeds enter restoration sites more slowly the more isolated they are. Kahmen & Poschlod (2004) and Olff, Pegtel & Vangroenendael (1994) also found functional traits to change along a successional gradient, although their studies did not compare the plant attributes of target communities.

relevance for the future restoration of calcareous grasslands

The extensive scope of this study has allowed us to examine whether restoration methods devised because of the results of small-scale local efforts or through ecological theory can be used with confidence when restoring recently abandoned arable land. Consequently, we have some recommendations for future restoration efforts of calcareous grassland, but it is an approach that could be equally beneficial to other areas of restoration ecology or of habitat management.

Much as it might be against the instincts of the countryside managers and policy-makers, these initial findings indicate that sites allowed to regenerate naturally are the most successful in terms of plant composition. Seeding and/or seed-mixtures used may actually delay the restoration process by inhibiting succession. These findings are likely to have masked the benefit of the best seed mixes and conditions under which they were sown, but they clearly emphasize the importance of devising the correct seed mixture for a site. Since the diverse seeding of restoration sites is an expensive process, both in seeds and in labour, a possible course of action would be to leave sites to naturally regenerate (with a management regime akin to that of reference sites) unless sufficient resources can be employed to ensure seeding is beneficial. This is not a new idea in restoration science; Prach (2003) has long been advocating ‘spontaneous succession’ on land that does not have especially adverse abiotic conditions. Unless there are ancient, good quality calcareous grassland sites within very close proximity, however, the lack of suitable propagules will be problematic.

We believe that the P:N ratio of ex-arable soils should be reduced until it is P that is limiting production and not N. Until that point, different plant species to those found in ancient calcareous grassland will be favoured, and the communities will differ. A range of strategies have been devised to alter the nutrient status of ex-arable soils (Marrs 1993; Walker et al. 2004). For instance, the addition of N to lower the P:N ratio (and in time lower the total P concentration, if combined with removing hay) could be extremely beneficial (Gough & Marrs 1990).

Acknowledgements

We are grateful to many landowners and managers who provided access to sites and information about them. Funding for this work was in the form of a NERC studentship awarded to Kate Fagan.

Ancillary