SEARCH

SEARCH BY CITATION

Keywords:

  • conservation management;
  • disturbance;
  • heathland;
  • safe-site limitation;
  • seed limitation;
  • state transition;
  • tree seedling recruitment

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  • 1
    Numerous studies describe thresholds at which transitions between alternate ecosystem states occur but few quantitatively delimit these conditions and present them in simple frameworks that are of use in ecosystem management. Previous research on heathland ecosystems has provided a site-specific statistical model that describes the determinants of the threshold point in heath–scrub vegetation transition, but the wider applicability, and thus utility, of this model was unknown.
  • 2
    Multi-site experimental manipulations were conducted to assess whether a consistent set of factors limited the recruitment of Betula species into heathland vegetation. Data were pooled and used to fit a single general statistical model. The applicability of this model to a wider range of environments was tested using two independent data sets.
  • 3
    The identity of the factors controlling Betula colonization, which include Betula seed bank density, phosphorus (P) availability and disturbance effects, were broadly similar between sites, but the strength of their effect varied widely.
  • 4
    The general model (explained deviance 59·8%) described Betula seedling densities as a function of biomass and necromass density, vegetation height, Betula seed bank density, P availability and soil water content. These relationships were complex, with numerous interaction and polynomial terms.
  • 5
    Although the model was derived from data from a single community type it was reasonably accurate in the prediction of seedling densities over a wider range of heath conditions. However, data capable of validating predictions of high seedling densities were not available.
  • 6
    Synthesis and applications. The apparent success of the general statistical model suggests that an approach incorporating multi-site experiments and statistical modelling can facilitate our understanding of ecosystem state transitions and inform the management of invasive species. By describing Betula invasion as a function of variables that are simple, general descriptions of the environment, the model can potentially inform management in a wide range of conditions. The results suggest that heaths close to seed sources and in the degenerate state of the dwarf shrub cycle are the most vulnerable to invasion, and management should target such sites as a priority. At regional scales, these conditions are probably most common in high soil phosphorus sorption capacity areas, where management should be prioritized.

Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Thresholds at which multi-species assemblies shift from one ‘stable’ state to another are an accepted part of theoretical and applied ecology (Scheffer et al. 2001; Suding, Gross & Houseman 2004). However, few experimental studies have addressed the critical points at which state transitions occur, thus hindering effective ecosystem management. State transitions are often determined by the population densities of ‘foundation’ species that, upon recruitment, engineer the new ecosystem state (Wilson & Agnew 1992; Jones, Lawton & Shachak 1997; Petraitis & Latham 1999). Thus characterization of the state transition threshold (transitional area) may be obtained by defining what determines the invasion of these species.

In this paper we describe an attempt to quantitatively delimit the transitional area of a heath–scrub vegetation change. Invasion of Betula spp. [both Betula pubescens (Ehrh.) and Betula pendula (Roth)] drives transitions between heath and scrub ecosystems (Mitchell et al. 1997, 1999; Hester, Miles & Gimingham 1991a,b) and threatens the conservation of lowland heath (Harrison 1976; Rose et al. 1999). Betula recruitment is influenced by factors related to both the availability of seed (seed limitation) and the germination and survival of seedlings (safe-site limitation). Many environmental factors, including vegetation density and phosphorus (P) availability, are potentially important axes of the safe site (Manning, Putwain & Webb 2004). However, deriving a general model of Betula invasion is complicated by the considerable landscape-scale heterogeneity of heathland ecosystems. For example, soil phosphorus sorption capacity (PSC) affects P availability (Manning 2002) and displays landscape-scale variation that correlates with patterns of invasion (Chapman, Rose & Basanta 1989).

The overall aim of this research was to extend the site-specific model of Manning, Putwain & Webb (2004) by deriving a general statistical model that quantitatively describes the conditions in which Betula invasion occurs. Such a model would be of greater utility to heathland managers than a detailed but site-specific model. We conducted an experiment identifying Betula recruitment limitations at three geographically distinct sites to assess the universality of the key determinants identified by Manning, Putwain & Webb (2004). Data from the experiments were pooled to formulate a general model describing Betula colonization as a function of relatively simple variables that are applicable to a range of lowland heath environments. It was decided a priori that this would only be attempted if some determinants were identified as universal. Data from independent sites was then used in an attempt to validate the general model. The results of these studies are discussed with reference to both scientific understanding of transitions between alternate ecosystem states and heathland management.

Methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

study sites

An experiment was replicated in three heathland regions with distinct patterns of Betula invasion and PSC type. PSC was initially hypothesized to control P limitation at each site but its relationship with available P has since been found to be more complex (Manning 2002). The regions were: (i) Dorset, a region with a slow rate of Betula invasion and low soil PSC (c. 70 µg P g−1); (ii) New Forest, where soil PSC is high (c. 1500 µg P g−1) but Betula invasion is minimal, a probable result of continuous long-term grazing by free-ranging animals; and (iii) Surrey, where Betula invasion is a severe problem but soil PSC is intermediate to the other regions (c. 700 µg P g−1).

Experimental sites within these regions were chosen to fit three criteria: (i) an Erica tetralix–Sphagnum compactum plant community (M16 of the UK National Vegetation Classification; Rodwell 1992) with some gaps 25 cm2 in the canopy (all sites contained an M16c community type, i.e. dominated by dwarf shrubs not Sphagnum spp.); (ii) close proximity to a linear patch of mature Betula; and (iii) flat topography. All three sites lie upon the Bagshot beds, Tertiary (Eocene) sandy deposits.

The first site (Dorset ) was located upon the Arne Peninsula, Dorset (UK national grid reference SY 976 894), and is described elsewhere (Manning, Putwain & Webb 2004). Soil PSC is low (231 µg P g−1and 323 µg P g−1 at 0–50 mm and 50–150 mm depths, respectively) but high soil organic matter (SOM) content results in similar P availability to the other sites (Manning 2002). The second site (New Forest), at Denny Wood, Hampshire (national grid reference SU 340062), is dominated by Erica tetralix (L.). Vegetation gap size is intermediate to the other sites. As the site is within the New Forest boundary a high grazing intensity was expected. The site is regularly burnt and soil PSC is the highest of the three sites (1463 µg P g−1 at 0–50 mm and 746 µg P g−1 at 50–150 mm). The third site (Surrey), at Horsell Common, Surrey (national grid reference TQ 014 598), is dominated by uneven-aged Calluna vulgaris ((L.) Hull), Molinia caerulea ((L.) Moench) and E. tetralix, and the vegetation gaps are the smallest of the sites. The site has been unmanaged for many years and the surrounding area has been heavily invaded by scrub, including Betula. PSC is intermediate to the other sites (1112 µg P g−1 at 0–50 mm and 251 µg P g−1 at 50–150 mm). Sites were cleared of Betula prior to initiating experiments.

experimental design

A randomized block design was used, with all plots (2 × 1-m) in each block approximately equidistant to the nearest seed source. At all sites, the first block was 13 m from the Betula strip. The other blocks ran parallel at distances of 16 m, 19 m and, at the Dorset site, 21 m. Three treatments, P addition, seed addition and disturbance, were applied factorially at all sites. The major difference in design between the sites was the number of treatment levels and replicates. The Dorset site had three seed addition levels [control (background only), background plus 50 viable seeds m−2 and background plus 250 viable seeds m−2], four P addition levels (control, one, two and three additions) and two disturbance levels (control, disturbance). There were two replicates per block and four blocks, equalling 192 plots. The smaller New Forest and Surrey sites had two seed addition levels (control and 250 m−2 levels), three P addition levels (control, one and three additions) and the same two levels of disturbance, with two replicates per block and three blocks, equalling 72 plots. There was a 50-cm guard row between plots and a 1-m guard row between blocks.

The P addition consisted of triple super phosphate applied at rates of 17·6 kg−1 P ha−1; approximately equivalent to that released in the burning of a 30-year-old-heath (Chapman 1967). The treatment was applied on three occasions: July 1999, January 2000 and June 2000. Seed addition occurred in early November 1999 and coincided with the end of natural seed rain. Seed-sowing rates were consistent with typical seed bank densities of Betula spp. (Thompson, Bakker & Bekker 1997). The disturbance treatment was applied manually with a mattock in June 1999. Further details of the treatments can be found elsewhere (Manning, Putwain & Webb 2004).

The effect of these treatments on long-term available P, as measured by ammonium oxalate extractable P (Pox), was studied in a subexperiment located between the plots of the main experiment. In this experiment, all P addition and disturbance treatments were applied to smaller 0·5 × 0·5-m plots. Timing and application rates were consistent with the main experiments. Soil samples of 0–50 mm depth were taken in November 2000 and analysed for Pox according to the method of Pote et al. (1996). At the Dorset site disturbance had no effect but P addition significantly increased P availability (two-way anova, F= 4·953,56, P < 0·01). At the Surrey site, P addition significantly increased P availability (F = 13·42,29, P < 0·001) but there was no disturbance effect on P availability. The New Forest site displayed dramatic between-treatment differences in P availability (F = 30·42,30, P < 0·001). Disturbance also had a significant effect (F = 5·81,30, P < 0·05) but there was no interaction between these variables.

variables

In November 1999 soil cores and seed bank samples were taken on regular sampling grids that maximized coverage of the sites. Thirty samples, located between the experimental plots, were collected at the New Forest and Surrey sites and 72 at the Dorset site. Soil of 0–50 mm depth was analysed for SOM content using the loss-on-ignition method, gravimetric water content and P availability (Pox). A bulked seed bank sample (five 30-mm deep cores with a total area of 98 cm2) was taken at each position of the sampling grids in February 2000. Samples were stored at 4 °C within 6 h of sampling, and seed bank estimation was conducted in a polythene tunnel between March and June 2000. Each sample was passed through a 5-mm sieve and spread onto compost in 0·2 × 0·3-m seed trays, placed alongside 12 controls in a stratified random pattern on a sand bed and irrigated continuously. Betula seedlings were removed upon identification. Species identity was not recorded as seedling-stage Betula cannot be differentiated (Atkinson 1992). Seed rain was measured at the Dorset site only (Manning, Putwain & Webb 2004).

Vegetation height and density and mineral substrate cover were recorded, using point quadrat methods, between July and September 1999 for the Dorset site and in September 1999 at the Surrey and New Forest sites. Betula seedling density was recorded when seedlings were approximately 1 year old (May 2001). Detailed descriptions of the methods used are given by Manning, Putwain & Webb (2004). Throughout this paper it is assumed that Betula seedling densities represent the likelihood of invasion.

analysis

All statistical analyses were conducted in S+ 6 for Windows (Anonymous 2001). Estimation of treatment effects was achieved by analysing Betula seedling density data with an analysis of deviance model with a log link and Poisson error. More detailed descriptions of Betula seedling densities at each site were achieved by converting manipulated variables into a continuous form and then combining these with covariate data to formulate general linear models with Poisson error and a log link. These models were fitted using the procedure described by Manning, Putwain & Webb (2004) and Crawley (2002). With this method, simplification from a maximal model, containing all possible terms, proceeds via stepwise deletion and re-insertion until a ‘minimum adequate’ model (MAM), containing only significant terms, remains. Significance of parameter deletion was assessed using likelihood ratio tests (LRT), using F as the test statistic to account for overdispersion. Interaction terms were restricted to the first order. Upon arrival at the MAM we deleted each term, noted the change in explained deviance and then re-inserted it. This allowed for measurement of the null deviance explained by each term, expressed here as deviance change on deletion (%DCD). Several preliminary data manipulations were required before fitting the MAM. Values for seed bank density and soil data were acquired by kriging, a process in which statistical models of spatial autocorrelation patterns are used in optimum interpolation (Rossi et al. 1992). Either a spherical or Gaussian variogram model was selected by best-fit procedures and fitted to the omnidirectional variogram of the P availability, SOM content, soil water content and seed bank data of each site. These models were then used in ordinary kriging to generate spatial grids of values. The means of two point estimates taken from each square metre of each 2 × 1-m plot were used as covariate data in the modelling process. Total seed bank density was estimated by adding estimates of background seed rain to the added seed rain at the Dorset site and by adding background seed bank density to added seed rain at the Surrey and New Forest sites. The means of background seed bank and seed rain at the Dorset site were remarkably similar (seed rain 25·75 m−2, SD ± 15·6; seed bank 26·14 m−2 SD ± 38·8), probably because sampling occurred shortly after the cessation of seed shed but prior to germination and seed bank persistence is uncommon in Betula spp. (Thompson, Bakker & Bekker 1997). Therefore, the seed bank to seed rain ratio was assumed to be 1. P availability was calculated by taking the mean P availability of the various P addition and disturbance treatment combinations, as estimated from the subexperiment, and adding the difference from mean background P availability levels, as estimated from the kriged grid. Disturbance was retained as a factor variable, alongside continuous variables: mean vegetation height, biomass density of the dominants, necromass density and mineral substrate cover.

formulation of a general model

The data from all three sites were pooled and used to fit a single general statistical model describing Betula seedling densities. The emphasis at this stage of the analysis was to describe relationships rather than test hypotheses. Although technically the data were not truly independent, they were treated as such for this model-fitting process and so caution must be applied when interpreting the significance terms. The pooled data set contained values for Betula seedling densities, total seed bank density, SOM content, soil water content, mineral substrate cover, mean vegetation height, P availability, density of the dominant species, necromass density and total biomass density. A preliminary inspection of these data using univariate and ordination methods was conducted in order to evaluate the degree of overlap between site conditions, in both single and multi-dimensional space. This was done to assess whether the model was fitted to data of multi-site origin. Ordination was performed using principal components analysis (PCA). Data were scaled to provide unit variance.

A single MAM, with a log link and Poisson error, was fitted using the procedure described above. The maximal model contained quadratic and cubic terms and interaction terms between polynomial and main effects, but not interactions between polynomial terms or higher order interactions. Interaction terms were removed first, then polynomial terms; main effects were removed last. Upon arrival at the MAM site-specific regression slopes were systematically added and removed for each term in the model and the change in explained deviance (ED) was noted. Significance of site-specific terms was assessed using LRT. The aim of this process was to assess the universality of the fitted relationships.

model validation

The validity of the general model was investigated using two independent data sets. Each contained the same variables as the fitted model measured on a 130-point grid covering 5 ha. The first site (Arne), at Arne Heath, Dorset (UK national grid reference SY 968890), is dominated by C. vulgaris and Pteridium aquilinum ((L.) Khun). Most of the vegetation is in the mature and degenerate phases of the dwarf shrub cycle (Watt 1947). Wetter areas containing E. tetralix and M. caerulea are found in places. Betula density in and around the site is low. The site is not grazed by livestock but sika deer Cervus nippon (Temminck) can be seen in large numbers nearby. The ecological history of the area is discussed by Pickess, Burgess & Evans (1992). The second site (Horsell Common), at Horsell Common, Surrey (national grid reference TQ 013612), is typical of many in the region in that traditional management ceased at an early date (before 1904) and there has been extensive scrub invasion in the last 50–100 years. The dominant species are C. vulgaris, E. tetralix and M. caerulea. Some areas of the site are regions of recent scrub removal; others are overlain with young trees.

Vegetation height and density and the abundance of seedling Betula were recorded using the methods of Manning, Putwain & Webb (2004). Vegetation surveys were conducted in July 2000 at the Arne site and in June 2001 at the Horsell Common site. Soil sampling was conducted at both sites in November 1999. Samples were analysed, using the same methods as for the experimental sites, for Pox, SOM, soil water content and seed bank density (the latter was sampled in February 2000).

Univariate and multivariate (PCA) comparisons of the experimental and model validation data sets were conducted in order to assess whether the experimental data covered the range of conditions present in the wider heathland environment. The validity of the general model was then tested by placing the data from the model validation sites within the general model to acquire predicted values to compare with actual seedling density values. As each model validation site was within 2 km of an experimental site, there may have been bias towards agreement with the general model. Betula seedlings in the cotyledon stage or with stem diameters 3·5 mm were excluded from the analysis; even favourable growth conditions are unlikely to produce seedlings larger than this in a single year. A number of data (n = 18) were removed from the Horsell Common data set. This was because of both a priori reasoning (e.g. plots coinciding with mature Betula and footpaths) and the loss of two soil samples.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

experimental results

There were large differences in the strength of safe site and seed limitation at the three experimental sites. At the Dorset site, seed limitation was the stronger factor although both P addition and disturbance had highly significant effects. The results are discussed in detail elsewhere (Manning, Putwain & Webb 2004). Seedling densities at the New Forest site revealed large, positive, disturbance (F = 89·41,58, P << 0·01, ED 30·1%) and P addition (F = 52·92,58, P << 0·01, ED 35·7%) effects and a small, but significant, interaction between these factors (F = 3·92,58, P < 0·05, ED 2·62%; Fig. 1a). The large P addition effect could be attributed to both safe-site limitation and large differences in available P between the treatment levels at this site. Background seed bank densities were high for all plots at this site (337·9 m−2, SD ± 169·2) and so seed addition effects were non-significant. Disturbance had highly significant positive effects on Betula seedling densities at the Surrey site (F = 13·61,58, P << 0·01, ED 13·34%; Fig. 1b) and P addition had a small positive effect (F = 3·82,58, P < 0·05, ED 7·45%). Seed addition effects were intermediate to the other sites (F = 8·61,58, P < 0·01, ED 8·41%), as were background seed bank densities (47·2 m−2, SD ± 32·65).

image

Figure 1. The effect of seed addition, disturbance and P-addition treatments on 1-year-old Betula seedling densities at the (a) New Forest and (b) Surrey sites. s1, control; s2, 250 seeds m−2; c, control; d, disturbance. The equivalent figure for the Dorset site is presented in Manning, Putwain & Webb (2004). Error bars represent the SEM.

Download figure to PowerPoint

site-specific statistical models

MAM revealed that seed and safe-site limitations operated at all sites but that the degree of limitation, and identity of the safe-site determinants, varied between sites. The Dorset site model (ED 69·2%) is presented elsewhere (Manning, Putwain & Webb 2004); total seed bank density was able to explain the most variation in Betula seedling densities, and vegetation densities and height and P availability were also important.

The New Forest site MAM (Table 1) (ED 83·2%) described seedling densities as a function of six variables, the most important of which was P availability (DCD 35·7%P << 0·01). The second largest contribution to ED came from M. caerulea density (DCD 6·40%, P << 0·01), which was positively associated with Betula seedling densities in control plots but proliferated in disturbed conditions to the exclusion of Betula (M. caerulea density × disturbance DCD 3·70%, P < 0·01). Negative effects of M. caerulea in disturbed plots were balanced, and often exceeded, by positive disturbance effects. These were better described by mineral substrate cover (DCD 2·72%, P < 0·01) than the disturbance factor. Mineral substrates were more abundant in disturbed plots; their effect was likely to evince a requirement for soil microsites free of root and algal competition. Calluna vulgaris density had a significant negative effect on seedling densities (DCD 3·71%, P < 0·01). The final variable included in the New Forest MAM was total seed bank density (DCD 2·42%, P < 0·01) that, although very high in most plots, could account for some variation.

Table 1.  Minimum adequate general linear model (MAM) (log link, Poisson error) describing 1-year-old Betula seedling densities at the New Forest site. Null model deviance, 276 on 71 d.f.; MAM deviance, 46 on 64 d.f.; explained deviance, 83·2%. Deleted terms: E. tetralix density, soil water content, SOM content, mean vegetation height, biomass density, necromass density, total vegetation density. DCD, % of null deviance change when deleted from MAM
VariableParameter valueStandard error of parameter estimateDCD P (F-test)
Calluna vulgaris density (hits m−2)−0·100·037 3·71< 0·01
Molinia caerulea density (hits m−2)   0·0390·037 6·40<< 0·01
Disturbance   3·271·155 3·80<< 0·01
P availability (Poxµg P g−1)   0·0070·0008635·70<< 0·01
Mineral substrate cover (%)   0·0590·021 2·72< 0·01
Total seed bank density (m−2)   0·00160·0006 2·42< 0·01
M. caerulea density × disturbance−0·120·0369 3·70< 0·01

The Surrey site MAM (Table 2) (ED 38·4%) had a poorer fit and less parameterization than the New Forest and Dorset models. Specific biomass terms did not improve model fit but total vegetation density (DCD 14·07%, P << 0·01) was retained. This effect was negative and may have represented light interception. Mean vegetation height also explained a considerable portion of the variance (DCD 13·25%, P << 0·01) and displayed a positive relationship with seedling densities. The disturbance term at this site (DCD 4·29%, P < 0·05%) probably represented the creation of small, competition-free, microsites. The size of the total seed bank density effect (DCD 6·25%, P < 0·05) was intermediate to the other two sites.

Table 2.  Minimum adequate general linear model (MAM) (log link, Poisson error) describing 1-year-old Betula seedling densities at the Surrey site. Null model deviance, 180 on 71 d.f.; MAM deviance, 111 on 67 d.f.; explained deviance, 38·4%. Deleted terms: all species-specific vegetation densities, soil water content (0–50 mm), SOM (0–50 mm), P availability, mineral substrate cover, biomass density, necromass density. DCD, % null deviance change when deleted from MAM
VariableParameter valueStandard error of parameter estimateDCD P (F-test)
Total vegetation density (hits m−2)−0·01750·003514·07<< 0·01
Disturbance   0·3850·143 4·29< 0·05
Mean vegetation height (cm)   0·2030·0413·25<< 0·01
Total seed bank density (m−2)   0·00260·0008 6·25< 0·05

general model

Univariate examination of the pooled data revealed considerable overlap in the values of all variables between the three sites with the exception of SOM content, which was much lower at the Surrey site, total seed bank density, which was much higher at the New Forest site, and P availability, which was lower at the Dorset site. PCA of the combined-site pooled data found that the experimental plots could be separated into distinct types (components 1–4 accounted for 67% of the variance) but these types were, in general, not specific to a particular site. Therefore, the model was based on data from > 1 site for most typical numeric combinations (Fig. 2).

image

Figure 2. Principal components analysis of the data used to fit the general statistical model. Component 1 accounts for 26% of the variance, 2 for 17%, 3 for 14% and 4 for 10%. Error bars represent 1 SD.

Download figure to PowerPoint

The general MAM fitted to the pooled data (ED 59·8%) contained six predictor variables (Table 3) but 15 terms, as there were numerous polynomial and interaction terms. The variables accounting for the most deviance were biomass density (DCD 35·25%, P << 0·01), an independent effect of mean vegetation height (DCD 20·48%, P << 0·01) which can be thought of as representing the size and frequency of gaps in the vegetation, total seed bank density (DCD 19·59%, P << 0·01) and P availability (DCD 9·15%, P << 0·01). Smaller effects were observed for necromass density (DCD 2·65%, P < 0·01) and soil water content (DCD 0·84%, P < 0·05). All other variables were deleted from the model.

Table 3.  General minimum adequate model (MAM) describing 1-year-old Betula seedling densities (m−2) in the M16 wet heath community (log link, Poisson error). Null model deviance, 1131 on 335 d.f.; MAM deviance, 535 on 320 d.f.; explained deviance. 59·8%. Deleted terms: species-specific vegetation terms, SOM content (0–50 mm) and mineral substrate cover. The data used in the modelling process were pooled from three experimental sites. The final two columns display the results of generality tests in which site-specific regression slopes were fitted for each term of the model after estimation of the MAM. DCD, % null deviance change when deleted from MAM. NS, P > 0·05
VariableParameter valueStandard error of parameter estimateDCD P (F-test)Fit improvement (% null dev.)Site- specificity test (P)
  • Quadratic term.

  • Cubic term.

Main effects
Soil water content (0–50 mm) (% total mass)3·23e-0029·67e-003 0·84< 0·051·16< 0·05
Necromass density (hits m−2)6·54e-0021·81e-002 2·65< 0·011·45< 0·05
Biomass density (hits m−2)−5·92e-0029·44e-00335·25<< 0·010·82NS
Mean vegetation height (cm)1·450·31e-00320·48<< 0·011·35< 0·05
Total seed bank density (m−2)−2·01e-0031·19e-00319·59<< 0·011·67< 0·01
P availability (Poxµg P g−1)2·35e-0024·27e-003 9·15<< 0·010·47NS
Polynomial terms
P availability−5·81e-0051·51e-005 3·22<< 0·010·32NS
P availability6·26e-0081·61e-008 1·08< 0·050·30NS
Mean vegetation height−6·81e-0021·37e-002 2·59<< 0·011·35< 0·05
Interactions
Biomass density × total seed bank density−8·28e-0051·11e-005 5·20<< 0·011·48< 0·05
Biomass density × mean vegetation height4·05e-0037·50e-004 2·13<< 0·010·94NS
Total seed bank density × mean vegetation height9·83e-0041·34e-004 4·70<< 0·011·79< 0·01
Mean vegetation height × P availability−8·64e-0072·22e-007 1·25< 0·010·30NS
Necromass density × mean vegetation height−1·1e-0023·07e-003 1·08< 0·051·62< 0·01
Necromass density × mean vegetation height4·45e-0041·27e-004 0·91< 0·051·55< 0·01

The fitted relationships were complex (Figs 3a,b, 4a,b and 5). When biomass and necromass density were low, mean vegetation height displayed a quadratic relationship with seedling densities, peaking in vegetation of intermediate height (11–14 cm). As vegetation density increased seedling densities declined, as did the curvature of their relationship with mean vegetation height, which had a small, positive effect where vegetation density was high. These relationships were more pronounced for living vegetation, a probable reflection of competitive influences (Fig. 3a,b). The model contained interactions between total seed bank density and both biomass density (interaction term DCD 5·20%, P << 0·01) and mean vegetation height (interaction term DCD 4·70%, P << 0·01); high Betula seedling densities only occurred where both seed and safe sites were non-limiting (Fig. 4a,b). These interactions probably resulted from the increasing likelihood of seed and safe site coinciding as seed became more abundant. Description of the relationship between Betula seedling density and P availability required quadratic (DCD 3·22%, P << 0·01) and cubic terms (DCD 1·08%, P < 0·05) and an interaction between P availability and mean vegetation height (DCD 2·59%, P << 0·01; Fig. 5). Where vegetation was short (mean vegetation height < 10 cm) and therefore, for a fixed density, more compact, Betula seedling densities were unresponsive to P availability; only at high P availabilities (Pox > 400 µg P g−1) did seedling density respond. Where mean vegetation height was > 10 cm, Betula seedling densities were highly responsive to P availability over the 0–200 µg P g−1 range.

image

Figure 3. The interactive effects of: (a) biomass density and mean vegetation height and (b) necromass density and mean vegetation height on the density of 1-year-old Betula seedlings as predicted by the general model (Table 3). All other variables in the general model were held constant at their mean in the pooled data set.

Download figure to PowerPoint

image

Figure 4. Interactions between safe site and seed limitation as estimated in the general model (Table 3). (a) Response of 1-year-old Betula seedling densities to total seed bank and biomass densities. (b) Response to total seed bank density and mean vegetation height. All other variables in the general model were held constant at their mean in the pooled data set.

Download figure to PowerPoint

image

Figure 5. The response of 1-year-old Betula seedling density to phosphorus availability (Poxµg g−1) and mean vegetation height as estimated by the general model (Table 3). All other variables in the general model were held constant at their mean in the pooled data set.

Download figure to PowerPoint

The fit of some terms was significantly improved by fitting site-specific regression slopes (Table 3). However, changes in ED were always < 2% of the null deviance, suggesting that most relationships were broadly similar at all sites. This suggested the fitted relationships were general but it must be acknowledged that a model-fitting procedure including site-specific terms from the offset would have resulted in a model of greatly different form.

model validation

Examination of the data revealed that conditions at the model validation sites varied over larger, or different, ranges to the experimental sites; only two of the six variables, biomass and necromass density, were adequately represented by the experimental data. Total seed bank densities and P availabilities were typically greater at the experimental sites [total seed bank density (m−2): Arne 1·81, SD ± 4·85; Horsell Common 74·4, SD ± 230·7; experimental 207·6, SD ± 193·7; Pox (µg g−1): Arne 48·4, SD ± 24·7; Horsell Common 45·2, SD ± 11·7; experimental 218·7, SD ± 133·8]. Soil water content was also greater, and less variable, within the experimental plots [soil water content (% total mass): Arne 55·7, SD ± 12·3; Horsell Common 45·0, SD ± 10·9; experimental 67·8, SD ± 4·4]. Mean vegetation height was greater at the model validation sites, particularly Arne, where tall C. vulgaris and P. aquilinum raised values considerably [mean vegetation height (cm): Arne 17·03, SD ± 5·50; Horsell Common 14·4, SD ± 5·47; experimental 10·64, SD ± 2·99]. The first PCA component accounted for 36% of the variance and distinguished between the wet, high-P, high-seed experimental plots and taller, low-P, low-seed bank dry conditions of the model validation plots. Components 2–4, which cumulatively accounted for 50% of the variance, made no clear distinction between the sites (Fig. 6).

image

Figure 6. Principal components analysis of the data from the model validation sites and the pooled data set of the experimental sites. All variables included in the general model were included in the analysis. Component 1 accounted for 36% of the variance, 2 for 19%, 3 for 19% and 4 for 12%. Error bars represent 1 SD.

Download figure to PowerPoint

Low Betula seedling densities at both model validation sites precluded formal testing of model predictions. Most predictions were for densities < 0·3 m−2 but the minimum positive actual value was 0·5 m−2, a consequence of quadrat size. However, the mean predicted and actual densities from each site were compared in order to calculate a coarse estimate of model accuracy. The mean predicted seedling density of the Horsell Common site was 0·10 m−2 (SE ± 0·02) and the mean actual seedling density was 0·11 m−2 (SE ± 0·05). At the Dorset site the mean predicted seedling density was 0·07 m−2 (SE ± 0·01) and the mean actual density was 0·00 m−2 (SE ± 0·00), i.e. no seedlings were recorded.

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

The results of this study reveal that Betula recruitment is controlled by both seed and safe-site limitation and that vegetation density, vegetation height and P availability can be key axes of the Betula safe site in a range of heathland environments. The findings extend and support the more detailed, but site-specific, results of Manning, Putwain & Webb (2004). The fitted relationships of the general model show that the transitional area of heathland ecosystems is defined by the coincidence of many favourable conditions; the non-optimality of a single important variable (e.g. vegetation density, seed bank density) may inhibit invasion.

The determinants of Betula colonization were broadly similar between sites but their relative importance varied widely. Variables accounting for differences in seedling densities were those varying over a critical range of values, for example seed bank densities at the Dorset site and P availability at the New Forest. Inconsistency in site-specific models may be explained by the use of separate variables representing similar phenomena, lack of within-site variation and site-specificity in the determinants. Seed availability explained seedling densities at all sites, and it probably limits invasion on many heaths, which are typically more isolated from seed sources. Overall, disturbance had a positive effect on seedling recruitment at all sites that operated via several mechanisms, the most important being changes to vegetation. However, disturbance effects, when isolated from vegetation effects, were inconsistent. Between-site differences in Betula response to P resulted from initial limitation and within-site variability in P availability. Low PSC resulted in fairly equal levels of P availability across the Dorset site, while at the New Forest site high PSC generated great differences between control and high-level P addition plots.

Seedling abundance patterns at the Surrey site were more stochastic than at the others. This stochasticity may be linked to the failure of plot-scale measurements to contain the fine-grained resolution that could account for infrequent coincidences of seed and safe site when both are strongly limiting. This concept is supported by the overdispersion of residuals in the models, including the general model, where predicted seedling densities were low.

By describing the constraints on recruitment of a species known to activate a switch between ecosystem states, the general model represents one the first attempts to describe the threshold of transition (e.g. the F2 point of Scheffer et al. 2001) in a general, empirical, quantitative and multi-dimensional fashion. Despite these achievements the model is probabilistic and does not identify a clear threshold between the two alternate states. Although it relies upon the assumption that seedling densities indicate the likelihood of transition, this is valid if the mortality of seedlings > 1 year old displays no density dependence.

The model suggests that low-density vegetation of intermediate height is the most invasible and possibly reflects both low light interception and protection from stresses such as herbivory and frost damage. Invasion is therefore most likely in the early building and degenerate phases of the dwarf shrub cycle and in mixed-species communities containing C. vulgaris damaged by the herbivorous heather beetle Lochmaea suturalis (Thoms.). The contrasting, uninvasible condition is dense and short vegetation, for example the closed canopies of the later building and mature phases. General failure of Betula to penetrate closed canopies is consistent with studies by Gong & Gimingham (1984), Miles (1974) and Marrs (1986). However, Gong & Gimingham (1984) found survival of seedling Betula in all ages of Calluna vegetation, thus supporting the proposal that low-density, effectively stochastic, invasion can occur in generally uninvasible conditions. Such invasion is unlikely to trigger vegetation shifts at larger scales but may increase the likelihood of future occurrence via raised seed bank densities.

The apparent success of the general model in predicting seedling densities at the model validation sites suggests that the identity and influence of predictor variables is constant over a reasonably wide range of heathland conditions. Its relatively accurate prediction was surprising as many predictions relied upon extrapolation of the fitted relationships. The most likely explanation for the apparent accuracy of these extrapolations is that the extrapolated variables had values resulting in low expected seedling densities. Tall vegetation heights, for instance, result in low predicted seedling densities. However, the extent to which these relationships remain constant over even greater spatiotemporal ranges of biotic, climatic and edaphic conditions is unknown, as is the model's capacity to predict accurately high Betula seedling densities. Minor inaccuracy of predictions at the low-seed availability Arne site also highlights inadequacy in the model; because seedling density is not restricted to seed bank density, seedling densities are overestimated in optimal safe-site plots with low seed input. It is also clear that the factors retained in the model, and their relative effect size, reflect not just their importance but also their variance within the study sites and during the year in question. The small effect size of soil water content, for instance, may reflect low variability at the study sites rather than its importance in the more variable range of natural heathland conditions. In summary, reasonable confidence can be held in the qualitative conclusions that can be drawn from the model but its value in quantitative prediction is yet to be fully confirmed.

In synthesizing the findings of this and previously published research it can be tentatively concluded that invasion is most likely where vegetation is sparse and of intermediate height, where propagule supply is plentiful, where P availability (Pox) is > 100 µg g−1, and where herbivores are absent; active management of such sites should be considered a priority by conservation bodies. It is important to note, however, that there is no single set of invasible conditions but rather a suite of combinations that occupies a narrow area of multi-dimensional space. Because non-optimality of a single factor can preclude invasion, managers seeking to halt invasion have several simple strategies available to them. Perhaps the most practical are tree removal and, where this is not possible, for example in invaded habitats utilized by nightjars Caprilmulgus europaeus (L.), the management for building-phase vegetation by mowing and regular burning. However, before detailed and fully informed management recommendations are made it is important to know which invasible conditions occur in unmanipulated heathland environments and how the identified factors vary at the larger spatial scales to which management is applied.

Management regimes and inherent regional differences, for example geology and climate, will alter the susceptibility to invasion. Regular burning, for example, returns the ecosystem to the heath stability domain, after its endogenous movement towards the vulnerable degenerate state, as it promotes nutrient loss, maintains short, dense vegetation and destroys Betula seed. Infrequent burning events, however, are likely to initiate invasion as dwarf shrub rootstocks are destroyed and large quantities of P are released (Bullock & Webb 1995). Mammalian herbivores also influence Betula invasion through both direct consumption and via impacts on nutrient cycling and vegetation structure. Browsing animals will have strong direct negative effects upon scrub colonists but may do little to influence physical properties. In contrast, grazing animals are less selective and so probably have smaller direct effects on Betula colonists (Bokdam & Gleichman 2000). Strong indirect effects of grazing animals may be expressed via high biomass consumption, which influences vegetation structure, nutrient supply rate and the spatial pattern of these determinants. The net balance of these influences may depend greatly upon the density, type and spatiotemporal distribution of the animals, but will, in most cases, be negative.

The large-scale correlation between soil PSC and scrub invasion (Chapman, Rose & Basanta 1989) may be explained by direct effects of PSC on P availability, but is more likely to represent indirect influences. Heathlands on low PSC soils have a low productivity that slows gap formation and P accumulation (Chapman, Rose & Clarke 1989). By implication a greater proportion of vegetation will be in the invasible degenerate state after management cessation in high PSC regions. A second implication is the raised probability that natural or accidental fires will occur during periods of the cycle, in which fire shifts conditions towards the transitional area. Severe burn events are particularly likely to trigger invasion in high PSC areas as large amounts of P may be retained within the topsoil. For these reasons we suggest that the intensity of management for short, compact vegetation and low P conditions, by grazing, mowing or burning, should be increased in high PSC regions.

As there has been no direct test we tentatively conclude that an interaction between management regimes and PSC controls heath–scrub transitions. This is best exemplified by the widespread invasion of the Surrey heaths (Harrison 1976). Early abandonment of traditional management appears to have interacted with high PSC to result in invasion that, by increasing the propagule density of the region, further increases the likelihood of invasion in remaining heaths. Dorset, in contrast, is typified by low PSC, later abandonment, more intensive management and, therefore, slower rates of invasion.

Petraitis & Latham (1999) present two models of ecosystem state transition, one in which the foundation species of the state alters the environment and out-competes that of the original state in a continuous process, and a second in which disturbance stimulates rapid transition to the alternate state. Transition to Betula scrub occurs via both mechanisms: invasion and exclusion of susceptible dwarf shrub vegetation via a positive feedback, and through severe disturbances including fires and beetle outbreaks. Thus heathland vegetation dynamics can be viewed as applicable to both the continuous and discontinuous models described by Briske, Fuhlendorf & Smeins (2003), even at a single, spatiotemporal scale.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

We thank the Royal Society for the Protection of Birds, English China Clays International, Horsell Common Preservation Society and Forest Enterprise for site access and the staff of CEH Dorset for their support. P. Manning was funded by a NERC/CASE CEH studentship.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  • Anonymous (2001) S-Plus 6 for Windows User's Guide. Insightful, Seattle, WA.
  • Atkinson, M.D. (1992) Biological flora of the British Isles. Betula pendula Roth (B. verrucosa Ehrh.) and B. pubescens Ehrh. Journal of Ecology, 80, 837870.
  • Bokdam, J. & Gleichman, J.M. (2000) Effects of grazing by free-ranging cattle on vegetation dynamics in a continental north-west European heathland. Journal of Applied Ecology, 37, 415431.
  • Briske, D.D., Fuhlendorf, S.D. & Smeins, F.E. (2003) Vegetation dynamics on rangelands: a critique of the current paradigms. Journal of Applied Ecology, 40, 601614.
  • Bullock, J.M. & Webb, N.R. (1995) Responses to severe fires in heathland mosaics in southern England. Biological Conservation, 73, 207214.
  • Chapman, S.B. (1967) Nutrient budgets for a dry heath ecosystem in the south of England. Journal of Ecology, 55, 677689.
  • Chapman, S.B., Rose, R.J. & Basanta, M. (1989) Phosphorus adsorption by soils from heathlands in southern England in relation to successional change. Journal of Applied Ecology, 26, 673680.
  • Chapman, S.B., Rose, R.J. & Clarke, R.T. (1989) A model of the phosphorus dynamics of Calluna heathland. Journal of Ecology, 77, 3548.
  • Crawley, M.J. (2002) Statistical Computing: An Introduction to Data Analysis Using S+. Wiley, Chichester, UK.
  • Gong, W.K. & Gimingham, C.H. (1984) Birch regeneration in heath vegetation. Proceedings of the Royal Society of Edinburgh, 85B, 7381.
  • Harrison, C.M. (1976) Heathland management in Surrey, England. Biological Conservation, 10, 21220.
  • Hester, A.J., Miles, J. & Gimingham, C.H. (1991a) Succession from heather moorland to birch woodland. I. Experimental alteration of specific environmental conditions in the field. Journal of Ecology, 79, 303315.
  • Hester, A.J., Miles, J. & Gimingham, C.H. (1991b) Succession from heather moorland to birch woodland. II. Growth and competition between Vaccinium myrtillus, Deschampsia flexuosa and Agrostis capillaris. Journal of Ecology, 79, 317328.
  • Jones, C.G., Lawton, J.H. & Shachak, M. (1997) Positive and negative effects of organisms as physical ecosystem engineers. Ecology, 78, 19461957.
  • Manning, P. (2002) Determinants of Betula spp. invasion of lowland heath. PhD Thesis. University of Liverpool, Liverpool, UK.
  • Manning, P., Putwain, P.D. & Webb, N.R. (2004) Identifying and modelling the determinants of woody plant invasion of lowland heath. Journal of Ecology, 92, 868881.
  • Marrs, R.H. (1986) The role of catastrophic death of Calluna in heathland dynamics. Vegetatio, 66, 109115.
  • Miles, J. (1974) Experimental establishment of new species from seed in callunetum in north-east Scotland. Journal of Ecology, 62, 675687.
  • Mitchell, R.J., Marrs, R.H., Le Duc, M.G. & Auld, M.H.D. (1997) A study of succession on lowland heaths in Dorset, southern England: changes in vegetation and soil chemical properties. Journal of Applied Ecology, 34, 14261444.
  • Mitchell, R.J., Marrs, R.H., Le Duc, M.G. & Auld, M.H.D. (1999) A study of the restoration of heathland on successional sites: changes in vegetation and soil chemical properties. Journal of Applied Ecology, 36, 770783.
  • Petraitis, P.S. & Latham, R.E. (1999) The importance of scale in testing the origins of alternative community states. Ecology, 80, 429442.
  • Pickess, B.P., Burgess, N.D. & Evans, C. (1992) The management of lowland heath from 1966 to 1989 at Arne nature reserve, Dorset, UK. Proceedings of the Dorset Natural History and Archaeological Society, 114, 227232.
  • Pote, D.H., Daniel, T.C., Sharpley, A.N., Moore, P.A., Edwards, D.R. & Nichols, D.J. (1996) Relating extractable soil phosphorus to phosphorus losses in runoff. Soil Science Society of America Journal, 60, 855859.
  • Rodwell, J.S. (1992) British Plant Communities. Volume 2. Mires and Heaths. Cambridge University Press, Cambridge, UK.
  • Rose, R.J., Webb, N.R., Clarke, R.T. & Traynor, C.H. (1999) Changes on the heathlands in Dorset, England, between 1987 and 1996. Biological Conservation, 93, 117125.
  • Rossi, R.E., Mualla, D.J., Jarne, A.G. & Franz, E.H. (1992) Geostatistical tools for modelling and interpreting ecological spatial dependence. Ecological Monographs, 62, 277314.
  • Scheffer, M., Carpenter, S., Foley, J.A., Folke, C. & Walker, B. (2001) Catastrophic shifts in ecosystems. Nature, 413, 591596.
  • Suding, K.N., Gross, K.L. & Houseman, G.R. (2004) Alternative states and positive feedbacks in restoration ecology. Trends in Ecology and Evolution, 19, 4653.
  • Thompson, K., Bakker, J.P. & Bekker, R.M. (1997) The Soil Seedbanks of North West Europe: Methodology, Density and Longevity. Cambridge University Press, Cambridge, UK.
  • Watt, A.S. (1947) Pattern and process in the plant community. Journal of Ecology, 35, 122.
  • Wilson, J.B. & Agnew, A.D.Q. (1992) Positive-feedback switches in plant communities. Advances in Ecological Research, 23, 263336.