PRE-SUBMISSION DRAFT, 06/04/2004

Detecting alternative stable states in long-term experiments to restore vegetation: an example with bracken infested Calluna moorland

M.G. LE DUC, C. TONG*, A. CHELU†, E. COX, R.J. PAKEMAN‡,

and R. H. MARRS

Applied Vegetation Dynamics Laboratory, School of Biological Sciences, University of Liverpool, PO Box 147, Liverpool, L69 3BX, UK; *Department of Ecology and Environmental Sciences, Inner Mongolia University, 010021 Huhhot, Inner Mongolia, China; †Department of Botany & Plant Ecology, USAMVB Timisoara, Calea Aradului 119, Timisoara, Romania; ‡Macaulay Land Use Research Institute, Craigiebuckler, Aberdeen, AB15 8QH, UK.

Correspondence author: M.G. Le Duc

tel. +44 (0)151 794 4775

fax +44 (0)151 794 4940

email

Running title: Experimental vegetation stable states

Summary

  1. In this work we apply current ecological theory on the nature of vegetation change (succession) following perturbation and parameter shift, when brought about by management intervention, to interpret field data from a long-term experiment.
  2. The experiment was established in the North Peak Environmentally Sensitive Area, Derbyshire, to examine the effect of vegetation restoration on land also subjected to Pteridium aquilinum control. It was designed to investigate sheep grazing and Calluna vulgaris seeding for restoration of heath, by incorporating the treatments at split-plot levels within P. aquilinum control treatments. The vegetation was monitored annually from 1993 to 2002.
  3. A multivariate ANOVA, using year as a split-split-split plot factor, was obtained using Redundancy Analysis (RDA) which showed that Calluna seeding on its own did not have a significant effect on the vegetation. However the four-way interaction term (bracken control x grazing x Calluna seeding x time) was significant.
  4. We used the constrained ordination from the RDA, in conjunction with standard-deviational ellipses, to examine the successional trajectories produce by the treatments. Three distinctly different trajectories (T1 – T3) were found.
  5. Trajectory T1 was produced by single perturbations such as herbicide application. The vegetation moved from a P. aquilinum –infested state to one with a reduced infestation, and then returned to the infested state, with perhaps increased P. aquilinum. T2 resulted from a number of different treatments in combination with stock-proof fencing. The vegetation was deflected from a direct return to the infested state through a stage dominated by Deschampsia flexuosa, but eventually returning to start conditions. T3 was produced by combination of annual cutting and light grazing. It resulted in a new stable state comprising acid-grassland species.
  6. It appeared that T2 may also eventually arrive at a new stable state, provided cutting is not part of the control regime. This was due to a number of well-established saplings that suggested Quercus sp.-Betula sp. woodland development.
  7. There was no indication that Calluna heath may be an alternative stable state under the treatment regimes tested. Nevertheless it was found that C. vulgaris could form a component part of the vegetation end points for trajectories T2 and T3.
  8. Ecological theory suggests that parameter changes might be necessary to overcome highly resilient vegetation states, as exemplified by P aquilinum infestation, and bring about alternative stable states. This has been shown to be the case at the North Peak experimental site.
  9. Synthesis and applications. The choice of target vegetation for restoration of a dense Pteridium aquilinum stand must be considered with care. There are likely to be only a small number of options, and these will require a significant amount of ongoing management intervention. It may be necessary to carry out a pilot study. At the study site, ongoing annual cutting together with light grazing will bring about an acid grassland sward.

Key words: acid grassland, attractor, Calluna seeding, grazing, perturbation, Pteridium aquilinum control, redundancy analysis, resilience, standard-deviational ellipses, successional trajectory
Introduction

Restoration of semi-natural communities such as grassland and heathland is not straightforward (Bakker & Berendse, 1999; Le Duc, et al. in press, a). Biotic constraints often exist through the highly resilient nature of the existing vegetation (van Diggelen & Marrs, 2003). The invasion of Pteridium aquilinum (nomenclature for higher plants - Stace, 1991; bryophytes – Hill, et al. 1991, 1992 & 1994; lichens – Coppins, 2002) produces such a situation. The species has a number of characteristics that produce a highly stable state. The annual frond canopy is very dense producing a 100% cover in the worst cases (Le Duc, et al. 2000). The fronds are deciduous, and a dense, persistent mat of litter results (Marrs, et al. in press). One major result is that the seed rain, seed bank and vegetation become decoupled (Ghorbani, et al. in press). Moreover, bracken has a very large underground rhizome system (c.20 to 50 t ha-1) that makes eradiction very difficult (Le Duc, et al. 2003), particularly in protected upland areas with difficult terrain.

As a result of the decoupling effect, vegetation existing under bracken is impoverished, and sometimes almost completely absent over considerable areas (Pakeman & Marrs, 1992). Further, the seed rain is often meagre. If this decoupling is sustained the change might be described as catastrophic (Scheffer, et al. 2001), since removing P. aquilinum from the site is unlikely to be followed by a return of the original vegetation. Thus, when the canopy is removed, hopefully for a period of a few years as part of a control programme, vegetation development can be very slow and unpredictable (Marrs, et al. 1998). An integrated restoration program should therefore include methods for encouraging target species, for instance by the application of appropriate seed and management (Anon, 1988).

Nevertheless, the outcome of restoration efforts from the apparently highly resilient (Beisner, et al. 2003) stable state of a Pteridium aquilinum (bracken) stand is often not predictable (Le Duc, et al. in press, a). The successional trajectories taken by the vegetation may not follow simple paths (Mitchell et al., 1997 & 1999; Suding et al., 2004).

CAUSES OF INFESTATION BY BRACKEN

Pteridium aquilinum infestation has become a serious problem in the UK, especially in the uplands and on marginal land (Pakeman, et al. 2000). As there are few recorded cases of successful establishment, in the wild, of the P. aquilinum from a sporophyte it must be assumed that under normal conditions all invasion is by vegetative spread from extant patches. Indeed this process has been measured, with rates of advancing front reported as c. 0.45 m yr-1 (Pakeman, et al. 2002). Spread from adjacent areas might occur as a result of relaxation of competition and other biotic pressures, or as a result of abiotic changes, perhaps through management activities or climate change (Bakker & Berendse, 1999; Suding, et al., 2004). There have been a number of changes in management activities in the UK uplands in recent years, with periods of intensification followed by the current policy for extensification (Haines-Young, et al. 2000). Thus the current problems with P. aquilinum infestation might be attributable to a number of interventions such as tree clearance, stocking reductions, and a reduced level of control (changed ecosystem parameters, sensu Beisner, at al. 2003, see below)..

THEORETICAL CONSIDERATIONS

Restoration treatments may be simple perturbations, or one-off events, that change the system’s state variables (species cover in this case). The vegetation will follow a successional trajectory towards an alternative state or attractor. If the resilience (Beisner, et al. 2003; Mitchell, et al. 2000) of the system is low, the perturbation may be enough to allow the successional trajectory to reach a new set of state variables, and thus a new stable state. However with high resilience, as is thought to be likely with P. aquilinum, the original state will rapidly re-establish (Marrs & Le Duc, 2000). An alternative approach would be to apply a continuing regime of altered management, that is, create a change to the ecosystem parameters (sensu Beisner et al., 2003). It has been common practice to use the former approach with restoration of P. aquilinum–infested sites, although changes in continuing management techniques are available as methods for controlling P. aquilinum, and some, such as regular cutting are in use.

Successional trajectories (Mitchell, et al. 2000) require varying input of effort and are probably finite in number, being constrained by attractors representing available vegetation stable states. Such states have a varying degree of stability, the more stable having the greater resilience. Such resilient vegetation may undergo moderate state changes without leaving the local basin of attraction (Beisner, et al. 2003). Thus in the case of P. aquilinum, we expect change to be limited by a number of possible trajectories requiring different amounts of effort. The most successful programme for restoration might be considered the one that optimises restoration effort with the resilience of the resulting vegetation, in an alternative stable state..

BRACKEN CONTROL

Bracken control is carried out extensively in the uplands, with Government grants available through a number of Agri-Environment schemes (Anon, 2002). The narrow spectrum herbicide, asulam, is licensed for aerial spraying, with c. 50 km2 yr-1 currently treated (Pakeman et al., in press), with variable success (Pakeman et al., 1998). This treatment is designed for a single application, perhaps followed a few years later by a second one.

A number of other technologies are available for bracken treatment (Pakeman, et al. 2000). Cutting, for example is often used where other vegetation present is high value, but this is often difficult in upland terrain. This may be applied as a continuing treatment with a number of annual cuts. A recent three-step herbicidal eradication process, using spot-spraying of asulam, has also been developed (Robinson, 2000). This requires ongoing annual treatment for a number of years.

VEGETATION RESTORATION

In many cases the loss of species from the system brought about by P. aquilinum infestation are such that it is essential to follow up bracken-control treatment with vegetation-restoration treatments (Marrs et al. 1998). The appropriate methods are likely to be specific to the target vegetation type. Appropriate seeding treatments might be accompanied by grazing control, or fertilizer application, etc. (Anon, 1988). Once again some such treatments may be applied as single perturbations, or others as ongoing interventions.

OBJECTIVES

The purpose of this work was to analyse the successional trajectories of vegetation, resulting from a long-term restoration experiment, in the context of current ideas in restoration ecology (Beisner, et al. 2003; Suding, et al. 2004). We hoped to test those ecological theories within a long-term, applied restoration experiment.

We used data from a long-term experiment in the UK uplands that had the objective to develop appropriate integrated methods encompassing bracken control and vegetation restoration towards a moorland or grassland objective (Le Duc, et al. 2000, 2003 & in press, a). From previous studies we know that dense P. aquilinum stands have high resistance and resilience in that there can be a rapid return to a stable state in some situations (Mitchell, et al., 2000). The aim from a restoration ecological view is to overcome the high resistance and move the vegetation to an alternative stable state with good resilience so that return to the old state does not occur.

It was hypothesised that different treatment combinations would result in different trajectories, which could be demonstrated using constrained ordination methods. The treatment combinations would have varying degrees of success in terms of alternative stable states. It was expected that one of these would be Calluna vulgaris (heather) moorland.
Methods

SITE

The experimental site is at Hordron Edge (141’W 5323’N, National Grid ref. SK 213 870) in the North Peak Environmentally Sensitive Area (ESA) within the Peak District National Park, Derbyshire, UK. It is 290 m a.s.l., with a general aspect of 275º and a 9º slope. Bracken infestation is substantial (Le Duc, et al. 2000) and dense litter beds are produced. The vegetation around the experiment is mainly upland acid grassland with much bracken infestation, but there are also areas of mire, woodland and heather moorland nearby (Tong et al., in press).

The experiment was set up in dense bracken with substantial quantities of litter covering most of the ground, but with isolated patches of other vegetation. The patches were mainly remnants of acid grassland, but one or two small pockets of heather moorland vegetation, with Calluna vulgaris and Erica tetralix, were also present. The objective was to test methods for returning the vegetation to Calluna moorland.

EXPERIMENT

This experiment, set up in 1993, was part of an extensive set of experiments established in four different locations within Great Britain. General descriptions of this scheme are given in Le Duc et al. (2000, 2003 & in press)

Experimental Design

A split-split plot experimental design was used with three blocks (replicates). The split-split design was necessitated by the nature of the cutting operations (main treatments) and stock fencing (sub-treatments). Each block was approximately 70 x 40 m, separated by at least 2 m, and each split-split plot was 10 x 5 m surrounded by a 1 m buffer zone.

Experimental Treatments

The main treatments, sub-treatments and sub-sub-treatments applied in the experiment are listed and coded in Table 1. The cutting operations were carried out using a Logic flail mower trailed by a Kawasaki 4x4 Quad, until 1999, and subsequently a hand operated AEBI model HC55 flail mower. All single cuts were carried out in June, with second cuts in August. The herbicide Asulox (Bayer plc) was applied by knapsack in August using 4.4 g active ingredient ha-1 in 400 litre water. Stock fencing was randomly allocated to the sub-plots, and was erected in early January 1994 six or seven weeks after seed application. The brash method for Calluna seeding used cut stems c. 20 cm long, applied at 13.2 t ha-1; the litter method used material sucked from under a mature Calluna stand, with a Muck Loada, (Euromec plc) and applied at 1.2 t ha-1. The source of the materials for Calluna seeding was about 3 km from the experiment, both methods included Agrostis castellana as a nurse crop, with seed applied at 12 kg ha-1, and took place in November 1993.

In total there are 36 three-way treatment-interaction terms. For ease of interpretation each is coded with a three-digit number (Table 1).

Sampling

Monitoring was carried out annually, from 1993 to 2002, in June and in August when bracken fronds were fully expanded. In 1993, before any treatment was applied, only plots destined for cutting that year were sampled so that undue trampling was avoided before herbicide spraying. From 1994 onwards all plots were sampled. Plots were sampled by randomly selecting, without replacement, the intersections of a 1-m grid. A 1-m quadrat was aligned with the grid with its bottom left-hand corner positioned at the selected grid point. Floristic measurements were made in June when all species, including bryophytes and lichens, were measured by estimating cover visually. In addition bracken litter depth was estimated, and Calluna seedlings counts (practical only in 1997 and 1998) were recorded. Estimates of the cover of bare ground, bracken litter, other litter and rock were also made. At the same time cover, density and length of bracken fronds were measured (Le Duc et al. 2000). In August only bracken frond measurements were carried out.

STATISTICAL ANALYSIS

Multivariate analysis

All multivariate analyses were carried out using the computer program CANOCO version 4.5 either in the WINDOWS version, or the console version for the most complex data sets. With low beta-diversity, and the assumption that many of the more abundant species would respond by expanding in cover after removal of the P. aquilinum cover, we adopted a linear model. Thus redundancy analysis (RDA) was used for further constrained analysis. Biplots were constructed using CANODRAW version 4 for WINDOWS (ter Braak and Šmilauer 2002).

Multivariate ANOVA

The randomization method described by ter Braak and Šmilauer (2002) was used to obtain an analysis of variance for the multivariate species data using RDA in CANOCO. As recommended by Legendre and Legendre (1998), we used 9999 Monte Carlo permutations for this analysis, with appropriate restrictions for the experimental design. For the multivariate analysis year was defined as a split-split-split plot factor (Mead 1990; ter Braak and Šmilauer 2002). The 1993 data, representing the experimental initial state, did not fully conform to the experimental design, and these data were included in the analysis as supplementary samples.

Standard-deviational ellipses

The dispersions of samples on the first two axes of the RDA ordination space were graphically represented using standard-deviational ellipses (SDEs). These were obtained using the algorithm ELLIPSE.SAS (Ricklefs and Nealen 1998) implemented in an EXCEL spreadsheet. The ellipse attributes size (L = SD(axis 1) x SD(axis 2), units of (log10[ordination units])2), shape (S = SD(axis 1) /SD(axis 2), dimensionless) and orientation (A = directional cosine of y, dimensionless), were calculated together with their standard errors (Ricklefs and Nealen, 1998). Finally Euclidean distances between ellipse centroids were also calculated (Le Duc et al. in press, b; Mitchell et al. 1999).

The ellipse attributes reflect a number of characters of the set of samples being investigated. Size is interpreted as the generality of the response, the larger the ellipse, the less the species composition of the samples of the defined set is limited by the environmental variables represented by the first two axes, and also implies a potential increase in species richness (dependent on the distribution and direction of species vectors in the biplot). The shape and orientation attributes together reflect the relative contribution of the environmental variables associated with each axis makes to the distribution of the samples of a particular set.
Results

SPECIES PRESENT

A detailed analysis of the individual species present, listed in Table 2, and their individual responses to treatment and through time are published elsewhere (Le Duc, et al. in press, a). Several trends are notable: the reduction in frequency of Agrostis capillaris (a species known to undergo a severe check following asulam application; Rhône-Poulenc 1997) with the onset of treatment, followed by a gradual return; an increasing presence of Calluna vulgaris from 1997 onward; the substantial increase in presence of Campylopus introflexus in 1997, followed by a slow decline; an increasing presence of Galium saxatile, checked in 2000; a big initial reduction in cover for Pteridium aquilinum (as measured in June, prior to full canopy expansion) after the start in treatment, followed by increasing cover to 2002 with the exception of a major check in 2000; the establishment of Rumex acetosella in 1998; an apparent reduction in the presence of Vaccinium spp. after treatment started.

RESPONSE OF BRACKEN TO TREATMENT

The effect of bracken-control treatments on frond cover of Pteridium aquilinum was significant, but slightly different, for June and August (see Le Duc, et al., in press, a, for a complete description of these results). Repeated measures analysis with polynomial contrasts (Le Duc, et al. in press, a) gave a 1st order (linear) effect with F(5,10) = 48.43** for June, and a 2nd order (binomial) effect for F(5,10) = 17.30** for August (significance levels include a Bonferroni adjustment for multiple testing – Cabin & Mitchell, 2000). The cut-twice-per-year treatment produces significantly the greatest reduction in cover from 1998 onwards for the August measurements, ending in 2002 with a cover of < 10% compared with all other treatments having > 30% and the experimental control having c. 100%. Results for June are not so obvious with the same treatment only emerging as best in 2001 (< 10% compared with control c. 60%).