Long-term effects of rotational prescribed burning and low-intensity sheep grazing on blanket-bog plant communities

Authors


Correspondence author. E-mail: calluna@liv.ac.uk

Summary

  1.  The importance of peatlands is being increasingly recognized internationally for both the conservation of biodiversity and the provision of ecosystem services; strategies are being developed world-wide to help maintain their integrity. Prescribed burning has been highlighted as a threat with considerable debate over its use as it is perceived to produce a Calluna vulgaris monoculture and a decline in preferred peat-forming species.
  2.  We investigated the impact of prescribed burning on vegetation composition and diversity in a long-term experiment at Moor House NNR in northern England. The study comprised a comparison between no-burn reference plots last burned in ca. 1924 and an experiment where all plots were burned in 1954/5. Within the experiment, the effects of very light sheep grazing vs. no grazing and three burning rotations (no-burn since 1954/5, repeat-burning at 10- and 20-year intervals) were tested.
  3.  Calluna vulgaris and Hypnum jutlandicum cover and bryophyte species richness increased in the least-disturbed, no-burn reference plots, but bryophyte cover did not. Lichen diversity declined.
  4.  Within the formal experiment, low-intensity sheep grazing had little impact but there were substantive changes produced by the different burning rotations. There was divergence between the burning rotation treatments with the least-disturbed, no-burn treatment changing towards a C. vulgarisH. jutlandicum community, whereas the most-disturbed 10-year rotation had a much greater abundance of both Eriophorum and Sphagnum spp.
  5.  Synthesis and applications. Our findings suggest that blanket-bog vegetation on peat responds to prescribed burning in a complex manner. Where burn return interval is long (>20 years), C. vulgaris becomes dominant and there was no evidence that preferred peat-forming species (Eriophorum/Sphagnum) increased. Where burn return interval is short (10 years), E. vaginatum/Sphagnum abundance increased. We found no evidence to suggest that prescribed burning was deleterious to the abundance of peat-forming species; indeed, it was found to favour them. These results inform conservation management policy for blanket bogs in the UK and more generally for future wildfire-mitigation strategies on dwarf-shrub-dominated peatlands elsewhere. Some lessons for the management of long-term experimental studies are also discussed.

Introduction

Moorland conservation on blanket bog is a major priority within the UK to maintain and enhance a number of ecosystem services, including carbon sequestration (Ostle et al. 2009), water provision (Yallop & Clutterbuck 2009), agriculture (Hester & Sydes 1992), sport and biodiversity (Littlewood et al. 2010). Central to this is the maintenance of the blanket bog in an active state, that is, continuing to accumulate peat (Bain et al. 2011). This is a main requirement for the sustainable conservation of these blanket bogs; in British systems, the main peat-forming species are Eriophorum and Sphagnum spp. (Bain et al. 2011). In many parts of northern England, the bryophyte cover (and the Sphagnum component in particular) has more or less disappeared as a result of combinations of past pollution (Tallis 1998; Caporn & Emmett 2009), overgrazing (Anderson & Yalden 1981) and wildfire (Albertson et al. 2009). Therefore, the conservation of Eriophorum and Sphagnum species is of major conservation interest.

One of the most contentious current issues in nature conservation in Great Britain is the prescribed burning of blanket bog (Bain et al. 2011). Such burning has been used to manage this vegetation for centuries, but has increased in the last 200 years or so for both sheep and grouse management (Hester & Sydes 1992). With the increasing recognition that upland blanket bogs play in providing a wide range of ecosystem services, the role of prescribed burning has been called into question (Bain et al. 2011). It is believed that this practice leads to vegetation dominated by Calluna vulgaris and that certain peat-forming species, and this includes Sphagnum species, are ‘fire sensitive’. A list of such species was collated by Littlewood et al. (2010), but they commented that the evidence was often vague with respect to any impact of prescribed burning. A systematic review of the topic (Stewart, Coles & Pullin 2005) found little evidence to show that prescribed burning damaged blanket bog. Nevertheless, prescribed burning is controlled by a voluntary code of practice that states ‘that there is a strong presumption against burning blanket bog’, although it may be implemented under specific circumstances (DEFRA 2007). One major defect identified by Stewart, Coles & Pullin (2005) was the lack of conclusive experimental evidence based on replicated, long-term experiments and specifically that there was almost no experimental evidence to guide the adoption of burning rotation lengths. Therefore, there is a need for experimental evidence of the impacts of prescribed burning of moorland vegetation on blanket bog.

In Great Britain in 1949, the Nature Conservancy was established and by 1952 had acquired the Moor House estate in the north of England as an experimental reserve. The Hard Hill Grazing and Burning Experiment was one of the first experiments to be established (1954/5) and was designed to test the effects of low-intensity grazing vs. no sheep grazing in combination with three prescribed burning rotations (a 10- and 20-year rotation plus a no-further-burn treatment). Some results have already been published from this experiment, but there are several drawbacks to these earlier analyses. For example, in the most comprehensive assessment of the impacts on vegetation (Rawes & Hobbs 1979), the statistical methods used to analyse the data are unclear, and at the time of the analysis, the 20-year and the no-burn treatment since 1954/5 were identical, and therefore, some of the published comparisons are spurious. Now, data are available to investigate the effects of the two sheep grazing treatments in combination with the three separate cycles, a 10-year cycle, a 20-year cycle and recovery after 47–48 years. We also compared results to unburned reference plots established adjacent to the experimental ones. Here, we use these experimental data to test two main hypotheses. Hypothesis 1, that plant species composition will change in the absence of prescribed burning; within this, there are two conflicting subhypotheses: (i) that species diversity will reduce through time (Harris et al. 2011) and (ii) that species diversity and especially diversity and/or abundance of Sphagnum species will increase (Littlewood et al. 2010). Hypothesis 2, that prescribed burning at different rotations and low-intensity grazing will affect both the abundance of individual species and plant community composition, and again, there are two subhypotheses: (i) long-rotation burning would favour Sphagnum spp. (Littlewood et al. 2010) and (ii) low-intensity grazing would favour C. vulgaris, lichens, but reduce the abundance of Eriophorum spp. (Rawes & Hobbs 1979).

As noted above, prescribed burning of moorland vegetation on peat is a cultural management practice associated with sheep and grouse production, which is particularly prevalent in Great Britain. The results from this study therefore apply at present just to a limited subset of current vegetation within in the boreal region. However, the principles derived from this study have direct relevance for informing future management of dwarf-shrub vegetation elsewhere in the boreal region. It is predicted that if global climate change occurs there will be with warmer and drier summers in northern latitudes, and this may result in increased wildfires (Albertson et al. 2010). Prescribed burning may be one technique that can be used to minimize damage in protected areas and around human settlements (Albertson et al. 2010). Wildfire extent can be substantive in the boreal region; between 1990 and 1992 for example, large wildfires in Alaska affected 2 × 106 ha of boreal forests, with many burns occurring over more than 20 000 ha (Kasischke & French 1995).

Materials and methods

The full experimental design and sampling methodology are detailed in Marrs et al. (1986).

Experimental Design

The experiment was set up on Hard Hill, Moor House in 1954/5 when four replicate moorland blocks (A-D), each 90 × 60 m, were burned along an elevational gradient (A = NY743330, 600 m; B = NY740330, 610 m; C = NY736330, 617 m; and D = NY738331, 632 m). Blocks A, B and D were burned in 1954 and block C in 1955. Within each block, two main-plot treatments (90 × 30 m) were allocated randomly, and these treatments were (i) sheep grazing (G = grazed) and (ii) no sheep grazing (E = enclosed). Within each main plot, three burning rotation treatments were allocated randomly to subplots (30 × 30 m), and these were (i) short-rotation burning (S, approximately every 10 years), (ii) long-rotation burning (L, approximately every 20 years) and (iii) unburned since year 1954/5 (N). Prescribed burning in the weather conditions prevailing at Moor House is very difficult, and in some years, burning is impossible; accordingly, burning timings could not be applied fully in accordance with the planned schedule and were applied as follows (Table 1a): 1954/5 (All), 1965 (S), 1975 (S & L), 1984 (S), 1995 (S & L). Thus, after the initial burn, the experimental data here represent the effects of four short-rotation burns, two long-rotation burns and recovery after 47–48 years (N). In addition to the formal experiment, each block had an associated unburned reference plot (denoted R, 10 × 10 m) set up outside the burn area delimited in 1954/5 but grazed by sheep. This vegetation was considered to have remained unburned for at least 30 years in 1954/5 (Rawes & Hobbs 1979) and possibly reflects a century of no burning management (pre–First Word War). The exact positions of the unburned controls were lost, but in 2011 these plots were relocated using a combination of map locations and aerial photography to ensure they were outside the original burn areas. We accept they may not be in identical positions, but they are extremely close (within metres), and given this, we assume they had remained unburned for 87 years at the time of the final sampling.

Table 1. History of (a) treatment application and (b) vegetation monitoring of the Hard Hill Grazing and Burning Experiment at Moor House National Nature Reserve. Key to burning treatments: S = short-rotation; L = long-rotation; N = no-burn since 1954/5; R = reference plots not burned in 1954/5. Note different sampling methods have been used, and in 1972/3, the sampling was split across 2 years. Domin = Domin scale and PQ = point-quadrat method
Main-plot treatments grazed/enclosedSubplot treatments burning rotationNo of burns since 1954Year of last burn appliedGrowing seasons between final sampling and last burn
(a)
EnclosedN11954/5547/48
EnclosedS519955
EnclosedL219955
GrazedN11954/5547/48
GrazedS519955
GrazedL219955
GrazedR01924Minimum 87
Part of the studyYear
196119651972197319731982 1991 2001a 2011
  1. a

    In 2001, one subplot in block A had one point-quadrat frame missing.

  2. b

    Because only one block was sampled, these data were not used in any analysis.

(b)
Reference plotsBlocks A–DA onlyb   A–D
Sampling method DominPQ   Domin
No. of quadrats/treatment subplot 5 × 1 m220 × 1 m2    
No of pins per quadrat  10 pins per frame   25 × 1 m2
Formal experimentBlocksA–DA–DAAB–DBlocks A–D 
TreatmentsAllAllL & S onlyN onlyAllAll 
Sampling methodDominDominPQPQPQPQ 
No. of quadrats/treatment subplot25 × 1 m25 × 1 m220 × 1 m220 × 1 m220 × 1 m220 × 1 m2 
No of pins/1-m2 quadrat  10555 
No of pins/1-m2 subplot  200100100100 
     Data from 1972/1973 were combined and denoted 1973; half the pins/quadrat were removed randomly from the 1972 data set to provide comparable sampling effort  

Vegetation Monitoring

The original aim was to monitor vegetation response approximately 7 years after each burning treatment was applied; this schedule has changed slightly as a result of delays in burning implementation (Table 1). In the first sampling (1961), the cover abundance of all species (vascular plants, bryophytes and lichens) plus litter, open water and bare ground present in 25 1 × 1 m quadrats randomly allocated within each subplot were assessed using the Domin scale, a subjective, semi-quantitative cover-abundance method (Currall 1987; Table 1b).

Between 1972/3 and 2001, the species composition within the formal experiment was recorded four times using point-quadrats (1-m-long frame with 10 pin positions at 10-cm intervals, pin diameter = 2 mm). Here, the central 14 × 7 m sampling zone in each subplot was delimited and 20 × 1 m2 quadrats were sampled randomly. Within each of these 20 quadrats, the point-quadrat frame was placed at five fixed spatial positions (Marrs et al. 1986); once fixed in position, one of the 10 pin positions within each frame was sampled at random (i.e. five pin counts per 1-m2 quadrat = 100 pins per subplot). In 1972, two pin positions per frame were sampled at each location (Table 1b), that is, n = 10 per 1-m2 quadrat; 50% of these pins were selected randomly for inclusion in the current analysis to maintain comparable sampling intensity. At each pin, the number of times a vascular plant touched the pin was counted, and bryophytes were counted as a single touch. To remove potential bias between these taxonomic groups, the point-quadrat data were expressed as the number of pin-presences per quadrat position (i.e. out of n = 5), that is, the additional information on vascular plant cover (multiple touches per pin) was removed.

The unburned reference subplots (R) have not been assessed over most of the period; indeed, data are only available on two occasions for all treatments (Table 1). In 1965, the vegetation was assessed in five randomly placed 1-m2 quadrats per plot (Table 1b), and in 2011, the vegetation in the relocated plots was assessed in 25 randomly placed 1-m2 quadrats per plot. On each occasion, vegetation was assessed using the Domin scale.

Data Preparation

The data set was voluminous and complex and required a substantive clean-up, first to bring species nomenclature to the same standard: Stace (2010) for vascular plants, Atherton, Bosanquet & Lawley (2010) for bryophytes and Dobson (2011) for lichens, and second to combine some groups that were recorded inconsistently. These changes are outlined along with variables measured (Tables S1 and S2, Supporting information).

Data Analysis

All analyses were performed using ‘nlme’ (linear mixed-effects models, LMM, Pinheiro et al. 2011) and ‘vegan’ for all multivariate analyses (Oksanen et al. 2011) within the R statistical environment (R Development Core Team 2012).

Hypothesis 1(a,b): in the absence of prescribed burning, species composition will change over time

The Domin cover-abundance data collected from the unburned reference plots in 1965 and 2011 were compared. Initially, the numbers of all species and of the different taxonomic groups (vascular plants, mosses, liverworts and lichens) for each of the four untreated plots were compared between years. Thereafter, overall species richness (number of species per quadrat), species richness by taxonomic group and diversity indices were calculated for each quadrat. The Domin cover-abundance scale data were converted to percentages using Currell's transformation (Cover (%) = (Domin score)2·6/4, Currall 1987) to make the scale approximately linear as a percentage and then logit-transformed as recommended for percentage data by Warton & Hui (2011). LMM were used to assess the differences between years (fixed factor) with blocks as the random factor for species richness variables, diversity indices and the cover of all species present in more than 25% of the quadrats. It is well known that an increase in sampling effort will influence species richness. Here, five quadrats per plot were sampled in 1965, whereas 25 quadrats per plot were sampled in 2011; this increase was designed to achieve a better approximation of species composition (see Fig. S1, Supporting information for species–area curves for these sites, Oksanen 2011). In order to eliminate the effects of different sampling effort, a simulation approximation was used; for all diversity measures (i.e. measures depending on richness), a resampling was carried out by selecting only one-fifth of the samples collected in 2011 by block, and then a linear mixed modelling was performed. This was repeated 999 times and a P-value density distribution composed (Robert & Casella 2009). Only density distributions that were significantly lower than < 0·05 are discussed. Permutational multivariate analysis of variance (PMAV) using Bray and Curtis distance was also used to examine and quantify the differences in floristic composition between the two sampling dates (Oksanen et al. 2011).

Hypothesis 2(a,b): prescribed burning at different rotations will affect the abundance of individual species including sphagnum

Here, point-quadrat data from the formal experiment in 1973, 1982, 1992 and 2001 were used; the total frequency of each species was calculated per subplot (n = 24 per year), and from this data set, species richness (overall and by taxonomic group) and diversity indices were calculated. LMM were then used to test for burning rotation and temporal effects (year × burn rotation interaction as fixed factors) on species richness and diversity indices as well as the frequency for all species present in more than 25% of the available subplots (n = 24). Frequency of litter, bare peat or soil and open water was also tested. LMM were performed for eight diversity measures, three environmental measures and species frequency (see Table S3, Supporting information). In LMM analyses, all frequency measures (%) were transformed (√n) to correct the problems with non-normally distributed errors and the random effects in the model were block/graze/subplot. The complete interaction model was tested, and where the interaction terms were not significant, they were deleted and simpler models assessed, that is, selecting the minimum adequate model (MAM; Crawley 2007).

Hypothesis 2(a,b): prescribed burning at different rotations will affect plant community composition

Here, point-quadrat data from the formal experiment in 1973, 1982, 1992 and 2001 as described above were used (n = 96, i.e. 4 years × 24 subplots). Initially, the relationship between the species community data was explored using detrended correspondence analysis (DCA, ‘decorana’ function, Oksanen 2011) and using Hellinger-transformed species frequency data as recommended by Peres-Neto et al. (2006). The DCA produced eigenvalues of 0·204, 0·150, 0·085 and 0·058 and gradient lengths of 2·052, 1·921, 1·460 and 1·151 for the first four axes, confirming that the linear model should be used. Therefore, the analysis was redone using redundancy analysis (RDA) fitted using the ‘rda’ function (Oksanen et al. 2011). The RDA model is discussed here as it is constrained on elapsed time since the start of the experiment. The correlation on environmental variables [frequency of (i) bare peat, (ii) litter, (iii) open water and (iv) an index of vegetation structure, i.e. the summed total of all species pin contacts per plot] was calculated using function ‘envfit’ (Oksanen et al. 2011) with 9999 permutations and plotted as passive variables on the ordination. The distribution of treatment effects was then visualized in ordination space as standard deviational bivariate ellipses (SD-ellipses), using the ‘ordiellipse’ function (Oksanen et al. 2011). Second, variation partitioning (Peres-Neto et al. 2006) was performed using ‘varpart’ function (Oksanen et al. 2011) to assess the relative contributions of grazing, burning and year in explaining the variation in species composition (Peres-Neto et al. 2006). Significance was assessed using a permutation test with the reduced model (n = 9999) stratified within year and block.

Lastly, principal response curves (PRCs, van den Brink & ter Braak 1999) were used to quantify and visualize the overall effects of burning and grazing combinations through time. PRC plots the temporal changes in species composition for each treatment as deviations from the control treatment represented as a zero line (here the most undisturbed enclosed, no-burn since 1954/5 plot; EN). Here, the transformed species data were standardized and the species scores scaled by eigenvalues, year was introduced as a covariable to control the overall temporal trend, and grazing × burning × year interactions were used as environmental variables. The model and first axis of the PRC were assessed using randomization tests with the reduced model and 999 permutations stratified within year and block. The species weights (bk) represent the affinity of each species with the treatments analysed: species with positive values increase with positive treatments, whereas species with negative values decrease, and species with values near zero do not show any response to treatments. This analysis was performed using the ‘prc’ function (Oksanen et al. 2011).

Results

Over the entire study, 83 species were recorded (see Table S4, Supporting information), of which 44 were found in both the experiment and the reference areas, and 10 and 29 species were only recorded in the reference plots and experiment, respectively. Four Sphagnum taxa were common to both the reference plots and the experiment, there were a further five Sphagnum species unique to the reference plots and four were unique to the experiment, thirteen taxa in all. Of particular note, Listera cordata was recorded in the reference plots for the first time in 2011.

Hypothesis 1(a,b): In the Absence of Prescribed Burning, Species Composition Will Change Over Time

There were substantial changes in the species composition in the reference plots (R) over the 47–48-year period (PMAV, R2 = 0·61, < 0·001). Eight species were found only in 1965, but 24 species were only found in 2011 (see Table S5, Supporting information). A greater number of mosses (17 new species) were recorded in 2011, especially Pleurozium schreberi, Rhytidiaelphus loreus and Sphagnum fallax. Other Sphagnum species were recorded at low frequency in 2011. Over time, there has been a significant increase in the species richness of mosses. However, there was no significant change in total species richness or evenness, but there was a significant decrease in lichen richness over time (Table 2). Only three species showed a significant change in cover between years, with C. vulgaris, Aulacomnium palustre and Plagiothecium undulatum increasing in cover (Table 2); all other species tested showed no significant change over time. Of particular note is the doubling of C. vulgaris cover.

Table 2. Change through time in the unburned reference plots within the Hard Hill Grazing and Burning Experiment at Moor House National Nature Reserve: measures that were significantly different through time (significant slope) detected using mixed-effects modelling. In the case of diversity measures, a randomization approach using LMM was used (9999 randomizations)
 Measure19652011Year effect (slope)
ArithmeticMean ± SEEstimatet-value
  1. Significance: *< 0·05; ***< 0·001.

DiversityMoss species richness (species/plot, log-transformed)2·60 ± 0·955·54 ± 0·800·0206·89***
Lichen species richness (species/plot, log-transformed)2·30 ± 0·970·39 ± 0·79−0·013−3·40**
Vascular plantsCalluna vulgaris (% cover, logit-transformed)26·8 ± 7·0055·3 ± 7·600·0238·26***
MossesAulacomnium palustre (% cover, logit-transformed)0·6 ± 1·033·5 ± 3·060·0022·29*
Plagiothecium undulatum (% cover, logit-transformed)2·0 ± 1·947·15 ± 3·970·0084·47***

Hypothesis 2(a,b): Prescribed Burning at Different Rotations Will Affect the Abundance of Individual Species Including Sphagnum

The MAM estimates are presented in Table S3 (Supporting information), and the arithmetic means (± SE) of the significant differences discussed here are presented in Table 3 (significant effects of burning) and Fig. 1 (significant year × burn rotation interactions).

Table 3. Arithmetic mean values (± SE) for those variables that produced significant change because of burning rotation treatment within the formal Hard Hill Grazing and Burning Experiment at Moor House NNR. Only species that produced significant burning rotation effects in the MAM derived from mixed-effects modelling (Table S3, Supporting information) are presented
 Burn rotationNo-burn since 1954/5(N)Short-rotation (10 years, S)Long-rotation (20 years, L)
  1. Different letters indicate significant differences between burning rotation treatments.

Diversity measuresSpecies number/subplot17·3 ± 0·8a16·6 ± 0·7a15·5 ± 0·7b
Number of moss spp./subplot6·1 ± 0·4a5·3 ± 0·4ab4·3 ± 0·4b
 Number of Sphagnum spp/subplot0·7 ± 0·1a1·1 ± 0·1b0·8 ± 0·1a
Total Sphagnum species5·9 ± 1·2a11·8 ± 2·3b3·8 ± 0·7a
Frequency measures (%)Bare peat0·8 ± 0·3a4·3 ± 1·0b3·9 ± 0·9b
Vascular plants Calluna vulgaris 83·7 ± 2·2a46·6 ± 5·0b59·9 ± 4·9c
Eriophorum vaginatum 41·2 ± 3·4a64·5 ± 2·3b57·8 ± 3·0c
Mosses Campylopus flexuosus 5·8 ± 1·4a22·0 ± 3·5b17·3 ± 2·5b
Sphagnum capillifolium 5·8 ± 1·2a11·6 ± 2·3b3·8 ± 0·7a
Figure 1.

Change in frequency of the four species that showed significant responses in the mixed-effects modelling (Table S3, Supporting information) to different year × burn treatment interactions through time at the Hard Hill Grazing and Burning Experiment at Moor House National Nature Reserve (1972–2001). Arithmetic means (± SE) are presented of frequency data; the enclosed, no-burn since 1954/5 was the intercept treatment.

Four variables (number of moss spp, number of Sphagnum spp, frequency of both all Sphagnum species and S. capillifolium) showed significant differences between burn rotations, but no year effect, indicating in the case of the Sphagnum-related variables greater number and frequency of Sphagnum species in the short-rotation treatment in comparison with no-burn and long-rotation (Table 3; see Table S6 (Supporting information) for Sphagnum species frequency of occurrence details). Moreover, the number of moss species was significantly lower in the long-rotation burning than in no-burn treatment, but short-rotation burning was not significantly different from the other two treatments (Table 3).

Five variables (species number, bare peat, C. vulgaris, Eriophorum vaginatum and Campylopus flexuosus) showed year and burn rotation effects (see Table S3, Supporting information). However, only the main species C. vulgaris showed an increasing trend over time for all burn rotations (slope = 0·032 ± 0·014), but with different intercepts for each treatment at 1973, suggesting significantly greater cover in the no-burn treatment (no-burn = 83·7 ± 2·2% vs. long-rotation = 59·9 ± 4·9% and short-rotation = 46·6 ± 5·0%). In contrast, the other five variables showed a decreasing trend with time (slope ranging from −0·001 to −0·065), and the significant differences between the three burning rotation intercepts reflected differential effects of burning rotation (Table S3, Supporting information), with E. vaginatum being greatest in the short-rotation burning, intermediate in the long-rotation burning and least in the no-burn treatment. Campylopus flexuosus and bare peat showed significantly lower values in no-burn treatment than in burning treatments, and S. capillifolium had a significantly low frequency in the no-burn treatment and long-rotation burning treatments and had the greatest frequency in the short-rotation burning (Table 3).

Three species showed significant year × burn rotation interactions (see Table S3, Supporting information, Fig. 1). Eriophorum angustifolium, Hypnum jutlandicum and Plagiomnium undulatum effectively showed similar responses with an increasing separation through time between (i) the no-burn since 1954/5 treatment and (ii) the short- and long-rotation burning. Eriophorum angustifolium increased with increasing burning frequency and both H. jutlandicum and P. undulatum increased in the no-burn since 1954/5 treatment over time; H. jutlandicum remained relatively constant in the short- and long-rotation burning, whereas P. undulatum increased slightly in these two treatments. In contrast, Empetrum nigrum showed more complex responses with a significant year × grazing × burn rotation interaction (see Table S3, Supporting information Fig. 1d). The intercept treatment (enclosed, no-burn since 1954/5) showed very little change through time, and both the grazed short- and long-rotation burning showed a small increasing response through time although there were fluctuations within the same range as the no-burn treatment. The main effect was in the grazed, no-burn since 1954/5 treatment where it was initially significantly greater, but reduced to similar values to the intercept by 1991.

Hypothesis (2a,b): Prescribed Burning at Different Rotations Will Affect Plant Community Composition

The first two axes of the RDA accounted for 22·1% and 10·8% of the variation (Fig. 2). Two of the four environmental variables increased negatively along axis 1, suggesting that there were greater frequencies of bare peat (r2 = 0·42, < 0·001) and vegetation density (r2 = 0·24, < 0·001) in the early phases of the study (Fig. 2a); frequency of litter and open water was not significant. The species plots show that the less-abundant species are centred near the origin (Fig. 2b) and the gradients reflect change in the major species (Fig. 2c). Axis 1 represents the temporal gradient with a change in species community composition from a liverwort-rich (−ve) vegetation to one dominated by higher plants and mosses (+ve). Axis 2 represents a gradation from vegetation dominated by E. angustifolium through E. vaginatum (+ve) to a dwarf-shrub-dominated C. vulgaris vegetation with H. jutlandicum and P. undulatum (−ve) (Fig. 2c).

Figure 2.

Biplots from a redundancy analysis of the point-quadrat data set (1972–2001) for the Hard Hill Grazing and Burning Experiment at Moor House National Nature Reserve constrained on elapsed time since the start of the monitoring (ET), illustrating: (a) all subplots and two significant environmental variables (BS = bare peat/soil, Struct = index of vegetation structure); (b) species plot; frequent = blue, less frequent = red; (c) the most frequent species; and (d–e) the standard deviational ellipses of the three grazing × burning rotation treatment interactions through time (red-empty = enclosed, blue-solid = sheep grazed); (d) = short-rotation (S), (e) = long-rotation (L) and (f) = no-burn since 1954/5. Species codes: Ap = Aulacomnium palustre, Ca = Calypogeia azurea, Cb = Cephalozia bicuspidata, Cfx = Campylopus flexuosus, Cls = Cladonia squamosa, Cm = Calypogeia muelleriana, Cv = Calluna vulgaris, Ds = Dicranum scoparium, Ea = Eriophorum angustifolium, En = Empetrum nigrum, Ev = Eriophorum vaginatum, Ga = Green algae, Hj = Hypnum jutlandicum, Kup = Kurzia pauciflora, Lv = Lophozia ventricosa, Mt = Mylia taylorii, Pn = Pohlia nutans, Ptu = Plagiothecium undulatum, Pu = Plagiomnium undulatum, Rc = Rubus chamaemorus, Sc = Sphagnum capillifolium. Species were designated frequent/less frequent based on the summed frequency of occurrence (SFOO) across all treatments and years: for frequent species SFOO ≥100; SFOO<100.

The grazing × burn rotation × year interactions (Fig. 2d,e,f) illustrated through the changing positions of the SD-ellipses show clear changes between burning rotations and very little difference between grazing treatments. There is an obvious separation between the least-disturbed treatments (N, no burning since 1954/5) and the short- and long-rotation burning (S & L). All treatments started with a relatively rich liverwort component, but as they have trended along axis 1, the least-disturbed treatments (EN & GN) have moved from a liverwort-rich vegetation through to one dominated by C. vulgaris. Both the short- and long-rotation burning have trended at a higher point on axis 2, suggesting communities with a greater abundance of E. vaginatum in the middle of the study and with an increasing amount of E. angustifolium in 2001.

These conclusions are borne out by the PRC analysis (Fig. 3); the overall PRC model accounted for 22·5% of the variation in species composition and the first PRC axis was also significant (999 permutations, < 0·001). The baseline is set to the most undisturbed treatment (enclosed, no-burn since 1954/5 treatment, EN) and is plotted as the zero line. All other treatment effects were positive and show that they all contain less C. vulgaris/H. jutlandicum (−ve) and more Eriophorum spp. (+ve) than the EN treatment. The grazed equivalent (grazed, no-burn since 1954/5 treatment, GN) stayed close to the baseline and there were no significant differences between them (= 0·771). In contrast, the other four treatments (EL, ES, GL and GS) moved away from the baseline through time showing maximum differences in 2011. These four treatment responses were associated positively (>0·5 bk) with a greater cover of E. angustifolium, E. vaginatum, C. introflexus and L. ventricosa through time, and negatively (<−0·5 bk) with the cover of A. palustre, C. vulgaris, P. schreberi and H. jutlandicum, which declined in abundance in these treatments relative to the EN treatment. There is therefore an increasing difference between the 10- and 20-year rotations and the no-burn since 1954/5 treatment; grazing has insignificant effects. The relative contributions of grazing, burning and year in explaining the variation in the species community data showed that 35·1% was explained by the three treatments (Pseudo-F = 13·86, < 0·01): grazing had a non-significant effect (1%, Pseudo-F = 1·64, > 0·05), but burning rotation and year accounted for 14·2% (Pseudo-F = 8·87, < 0·01) and 19·3% (Pseudo-F = 23·71, < 0·01), respectively, with very little shared variation.

Figure 3.

Principal response curves illustrating the relative changes in plant community performance through time at the Hard Hill Grazing and Burning Experiment at Moor House National Nature Reserve. The change in species is relative to the first axis of a RDA, and the species position on this axis is represented on the vertical axis. See Fig. 2 for species codes.

Discussion

Prescribed burning of blanket bog is a contentious issue (Bain et al. 2011), and the debate is hindered by the lack of long-term experimental data (Stewart, Coles & Pullin 2005). The Hard Hill Grazing and Burning Experiment set up at Moor House in 1954/5 is the only one that has tested long-term effects of prescribed burning. Unfortunately, because of the way the experiment has been run, there were four problems for data interpretation: (i) there was no pretreatment vegetation assessment and hence, initial within-block homogeneity was assumed; (ii) two recording methods were used; (iii) although a balanced experimental design was used within the formal experiment, the reference plots were outside this design; and (iv) the reference plots were only sampled occasionally. The lessons for running long-term experiments are obvious: derive a statistically rigorous experimental design, do not compromise by missing out ‘unnecessary’ treatments and use a standard recording method throughout. Nevertheless, the results showed that vegetation change does occur if the habitat is left unburned. There were compositional changes in all the treatments and reference sites, but surprisingly, the least-disturbed treatments (no-burn treatment) and the reference plots showed the greatest C. vulgaris increases and abundances, and short- and long-rotations prescribed burning reduced C. vulgaris abundance favouring Eriophorum spp. and Sphagnum-related variables. Thus, for Hypothesis 1a, it appears that no burning favoured C. vulgaris although overall species richness was not reduced here; thus, the overall conclusion of Harris et al. (2011) for species diversity derived for a study on very productive moorland is not supported here. However, the results also showed that prescribed burning at different frequencies does affect plant species composition in different ways: the different burning rotations changed both individual species abundance and community composition (Hypothesis 2a,b, accepted). But surprisingly, the hypothesis that no burning would enhance Sphagnum spp. and other peat-forming species (Hypothesis 1b) is not supported (Littlewood et al. 2010).

Effects of No-Burning Regime – Changes in the Reference Plots

The reference plots have not been burned since at least 1923 (Rawes & Hobbs 1979), but our study measured change in composition between 1965 and 2011 (46 years). During this time, there have been substantial changes in species composition, with increases in moss species richness and a reduction in lichen species richness. One positive result was the first recording of Listera cordata (locally rare) within the reference plots, although we cannot ascertain whether this is a real increase or a result of the increased sampling effort; this species is an indicator species of the C. vulgaris–E. vaginatum mire community in Britain (M19, Rodwell 1991). Further, long-term monitoring is needed to clarify this result. However, a negative result is that the C. vulgaris cover doubled during the sampling period dominating the community. The reason for the loss of lichen species remains unknown but one hypothesis is that lichens cannot persist as the abundance of C. vulgaris increases, supporting at least for this life-history category Hypothesis 1a [based on Harris et al. (2011)]. However, the overall bryophyte richness showed the opposite response. Thus, a no-burning regime affects different taxonomic groups differently.

The compositional changes were produced mainly because a greater number of moss species were detected in 2011, especially Pleurozium schreberi, Rhytidiadelphus loreus and Sphagnum fallax, none of which were detected in 1965 but all had a relatively high frequency in 2011. It should be noted that there was no significant increase in overall moss or Sphagnum cover, and the cover of only two mosses (A. palustre and P. undulatum) increased significantly. These changes may have been caused by a combination of (i) successional process towards C. vulgaris-dominated community, (ii) an increased sampling intensity that produced a better measure of species richness and (iii) a reduced pollutant load. The increased sampling intensity in the latest survey was deliberate and based on two issues: (i) the species–area curves indicated that increasing the sampling effort from 5 to 25 subsamples/plots increased the accuracy of species composition estimates in the reference community and (ii) other long-term experiments at Moor House had indicated a considerable bryophyte species reduction where litter accumulation was greatest (ungrazed plots, Marrs, Bravington & Rawes 1988) and this was considered a potential damaging factor in the reference plots.

This vegetation change must also be set in the context of impacts of aerial pollution. The moors in the southern Pennines have very low species diversity, and they are specifically devoid of typical blanket-bog bryophytes including Sphagnum spp. (Caporn & Emmett 2009), ascribed to past aerial pollution (Tallis 1998). The atmospheric pollutant loads are now reducing and it is expected that some populations of affected species might recover. The flora at Moor House, which is the northern Pennines and subject to a lower pollutant load, retained a bryophyte component throughout the twentieth century so pollutant impacts were presumably less than those further south. However, it does not mean that there was no effect; there has probably been a less damaging effect. The changes to the bryophyte flora noted here may be, at least in part, caused by a reduction in these pollutant inputs. As example, the frequency of Sphagnum fallax, a pioneer species resistant to pollution (Ferguson, Lee & Bell 1978), increased the most, other Sphagnum spp. being relatively infrequent.

From a conservation point of view, the species compositional change brought about by no burning for at least 87 years is both positive (increased Listera cordata (albeit requiring confirmation) and bryophytes species richness) and negative (increased C. vulgaris, reduced lichens, no-increase in Sphagnum cover).

Effects of Sheep Grazing and Different Burning Rotations

Sheep grazing had very little effect on vegetation composition and diversity, and this is in contrast to the conclusions of Rawes & Hobbs (1979) who suggest that this low-intensity grazing treatment (albeit with no burning) would be acceptable for conservation; this conclusion is not borne out by the longer-term data presented here. Thus, Hypothesis 2a,b with respect to sheep grazing is rejected. This is not all that unexpected because the summer-only sheep grazing intensity is very low (0·1–0·3 sheep ha−1 on Hard Hill, Rawes & Welch 1969). The grazing level is lower than those reported for wet heath (Hulme et al. 2002), but unfortunately there are no other experimental grazing studies on comparable Calluna-dominated blanket bog.

Our results indicate that prescribed burning has important effects on the moorland community; hence, Hypothesis 2(a,b) is accepted with respect to burning rotation. It is often suggested that prescribed burning promotes monocultures of C. vulgaris and reduces Sphagnum abundance (Littlewood et al. 2010); the opposing hypothesis must be that no burning will favour a reduction in C. vulgaris and an increased Sphagnum abundance. None of these suggestions were supported here; indeed, the least-disturbed treatments (reference and no-burn since 1954/5 treatment) had the greatest C. vulgaris abundance, and moreover, it has increased steadily over the last 47–48 years. This appears to be associated with an increase in H. jutlandicum, a result similar to that reported for postfire chronosequences on moorland elsewhere in upland Britain (Burch 2008). Whilst it is possible that C. vulgaris will reduce and be replaced in the longer time under a no-burn scenario, this would have to occur after a period of more than 87 years at this site.

There was no change in Sphagnum species abundance through time, but there were only changes between the burning rotation treatments. The greatest frequency of the most common Sphagnum species (S. capillifolium) was in plots burned every 10 years, exactly the opposite of expectation. This was also true for Eriophorum spp. These results suggest that an open canopy with reduced C. vulgaris as a result of prescribed burning will favour Sphagnum colonization and growth (more bare ground and little competition) than longer rotations. Indeed, it seems that if Sphagnum species were relatively uncommon at a site, short-rotation burning might produce conditions to encourage Sphagnum recolonization. However, the impact of short- vs. long-rotational burning on moors with a higher Sphagnum abundance might be different from those presented here. Nevertheless, the vegetation with the lowest abundance of C. vulgaris and greatest Eriphiorum spp. and S. capillifolium, a preferred option for many conservation bodies managing blanket bog (Bain et al. 2011), was achieved on the most-disturbed, short-rotation burning (every 10 years).

In both the burning experiment and reference plots, S. capillifolium was the most frequent Sphagnum species, but it showed an increasing frequency within short-rotation burning; other Sphagnum species were detected at low frequency, and then, they have appeared and disappeared over the time course of the experiment with no apparent relationship to treatment. There was certainly no evidence that prescribed burning hindered Sphagnum dynamics. This suggests that there is a background abundance of S. capillifolium across the site, but other Sphagnum species colonize and become extinct through time. Alternatively, it is possible that these other species are present at such low frequency that the variability in their occurrence is an artefact of the point-quadrat sampling technique. This remains to be tested using more detailed observations.

A negative result is that the frequency of bare peat increased with burning frequency. However, this does not indicate substantive patches of bare peat within the plot, but rather a reduced frequency of litter and bryophytes at the peat surface. It is of course also important to realize that here only the species composition of the vegetation has been considered, and other ecosystem functions are also affected by prescribed fire, including plant growth and peat-forming processes. It is noteworthy that three moss species (H. jutlandicum, P. undulatum and S. capillifolium) have recently been shown to show inter- and intraspecific differences in short-term carbon cycling (Orwin & Ostle 2012). These must also be considered in developing management plans for moorlands on blanket bog.

Management Implications

The evidence reported here suggests that leaving blanket bog on a very long prescribed burning rotation at this site will increase the Calluna vulgaris–Hypnum jutlandicum component in a similar manner to the reference plots. Where a short-rotation prescribed burning programme is used, the vegetation will shift towards one dominated by peat-forming Eriphiorum spp. and Sphagnum here. This may be because the fires applied here resemble ‘cool burns’ (Harris et al. 2011) because the vegetation is always relatively moist because of the high rainfall. With a ‘cool burn’, relatively little damage to the underlying moss layer might be expected, but this needs to be confirmed by experiment. At present, this research is applicable for the management of moorland managed for conservation purposes or where the management is for sheep and red grouse. In future, with predicted climate change, wildfire in these fire-adapted ecosystems is expected to increase, and prescribed burning may be one solution to minimize long-term, ecological damage. Further research is needed, however, to derive optimal prescribed burning strategies for the mitigation of such future wildfire-induced damage.

Acknowledgements

This work would not have been possible without the foresight and persistence of staff of the Nature Conservancy and its successor bodies and the UK Environmental Change Network. In addition, we thank the BiodivERsA FIREMAN program (NERC/Defra), the Ecological Continuity Trust, the Heather Trust and the Basque-Country Government (JGA BFI-2010-245) for financial support. Mrs S Mather and Ms S Yee prepared the figures. RHM thanks Professor Bill Heal and the late Michael Rawes for enthusing him in the importance of long-term, manipulative experiments.

Ancillary