Matrix modelling of prescribed burning in Calluna vulgaris-dominated moorland: short burning rotations minimize carbon loss at increased wildfire frequencies

Authors


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

Summary

  1. Moorlands store large amounts of carbon providing a valuable ecosystem service. In the UK, prescribed burning is often used to manage moorlands, which can produce both positive (biodiversity enhancement and wildfire prevention) and negative impacts (carbon release and reduction in some ecosystem services provision). This issue is pertinent, as wildfire incidence may increase under climate warming, increasing damage to both conservation value and ecosystem services. To better manage these fire-prone ecosystems, an understanding of the effects of prescribed-burning regimes and wildfire on moorland fuel loads is required. As a first approximation of the relative impacts of prescribed burning and wildfire, we have modelled above-ground fuel-load accumulation and carbon release under varying wildfire return intervals at a study site in the Peak District, UK.

  2. Above-ground fuel-load accumulation following prescribed burning was assessed using a chronosequence study and combined with carbon loss measurements from prescribed-burning experiments. The stable age structure of vegetation under varying prescribed-burning rotations was then predicted using an age-structured matrix model and moorland above-ground fuel load calculated using a boot-strapping approach. Finally, long-term carbon losses were predicted under varying wildfire return intervals. Delayed vegetation regeneration was also modelled. The model does not consider below-ground carbon or the effect of prescribed fire on wildfire probability.

  3. There was a clear interaction between prescribed-burning rotation interval and wildfire return interval. At 50- and 100-year wildfire return intervals, carbon losses were minimized by short prescribed-burning rotations. However, under a 200-year wildfire return interval, carbon loss was minimized by long rotation intervals where delayed regeneration was modelled.

  4. Under a 50-year wildfire return interval, 8-year prescribed-burning rotation intervals could reduce carbon loss by 22% or 34% compared with 25- and 50-year rotations, respectively.

  5. Synthesis and applications. The modelling approach outlined here provides a first approximation to the above-ground carbon balance between prescribed burning and wildfire frequency at a single site. This may be useful in other dwarf-shrub-dominated ecosystems if prescribed burning is to be used to mitigate the effects of wildfire. At our study site, long prescribed-burning rotations may minimize carbon loss at low wildfire return intervals. However, if wildfire incidence increases, more frequent prescribed burning is likely to minimize overall carbon loss. Well-informed prescribed burning on a short rotation may produce smaller carbon losses than longer rotations under future climate conditions.

Introduction

UK moorlands store in the region of 3000 Mt carbon (SEERAD 2007) which, combined with moorlands world-wide, represents a significant global carbon sink. In the UK, moorlands are cultural landscapes, created and maintained by human actions to prevent succession to woodland, and dominated by dwarf shrubs, particularly Calluna vulgaris (L.) Hull. intermixed with Eriophorum angustifolium Honck., Evaginatum L. and Sphagnum spp. L. (hereafter referred to by generic names unless in community descriptions), often growing on peat (>50 cm depth). Large areas can be described as C. vulgarisE. vaginatum blanket mire or E. vaginatum blanket and raised mire (M19/M20) (Rodwell 1991). Prescribed fire has been used for centuries in the UK to manage Calluna-dominated moorland for grazing and, more recently, for red grouse Lagopus lagopus scoticus Latham production (Gimingham 1972). Prescribed burning is carried out from late autumn to early spring and aims to achieve ‘light’ fire severity (sensu Keeley 2009), whereby surface litter, mosses and shrubs are charred or consumed, but the soil organic layer remains largely intact (Davies et al. 2010). Wildfires, however, are mostly accidental or caused by arson (Albertson et al. 2010) and tend to occur in spring or summer (Legg et al. 2007). They may cover large areas and burn with far greater intensity and severity, sometimes consuming all above-ground fuel load and significant amounts of underlying peat (Maltby, Legg & Proctor 1990). Inevitably, all fires release carbon into the atmosphere and where peat is burned carbon losses could be considerable. Current predictions suggest that summers will become hotter and drier due to climate change (Jenkins et al. 2009) causing moorland wildfire frequencies to increase (Nilsen, Johansen & Velle 2005; Albertson et al. 2010). However, despite the potential increase in wildfire risk and the need to conserve moorlands within the UK, especially those on peat soils, relatively little is known about the impact of prescribed burning on moorland carbon balance.

Calluna is a fire-adapted species and shows increased regeneration from seed in response to regular fire, provided that temperatures and exposure times do not exceed lethal values (e.g. Nilsen, Johansen & Velle 2005; Måren et al. 2010; Velle, Nilsen & Vandvik 2012). Following burning, Calluna can also regenerate by vegetative means, which may be more rapid than seed regeneration, but is reduced when plants older than 15–20 years are burned (e.g. Miller & Miles 1970; Hobbs & Gimingham 1984b; Davies et al. 2010) and may be absent in some regions, for example northern European heathlands (Nilsen, Johansen & Velle 2005; Velle, Nilsen & Vandvik 2012) and parts of lowland Britain (Marrs 1986). Where vegetative regeneration is suppressed, regeneration from seed may result in a lag period of little or no growth before germination and establishment (Grant 1968), although other studies have observed seed regeneration within a year of burning (e.g. Miller & Miles 1970; Nilsen, Johansen & Velle 2005).

Since a recent review of prescribed burning (Tucker 2003), there has been considerable debate over optimum burning regimes and whether the practice should be halted altogether. Moorlands provide a range of ecosystem services (e.g. carbon sequestration, water provision, flood protection, recreation), and, with an increasing policy focus on ecosystem service provision, this creates conflict over which services should take priority and the resulting management implications (Marrs et al. 2007). Carbon sequestration by moorlands is particularly pertinent to the debate over burning. On the one hand, prescribed burning releases carbon to the atmosphere and may contribute to other carbon losses such as increased levels of dissolved organic carbon (DOC; Yallop & Clutterbuck 2009) although evidence for the latter remains unclear (Holden et al. 2012). On the other hand, there is some evidence that prescribed fire can provide some reduction in area of wildfire (leverage) in some ecosystems (Narayan et al. 2007; Vilén & Fernandes 2011), primarily through reduced fuel load, although it may also provide fire breaks to impede wildfire spread (Costigan et al. 2005). Conversely, Bradstock et al. (2012) showed prescribed fire produced negligible benefits. Thus, prescribed burning might constitute an acceptable carbon loss when set against the greater losses, it could prevent.

Despite the increasing demand for evidence-based conservation and management, reliable predictive information to inform policy is often lacking (Stewart, Coles & Pullin 2005), and little is known about minimizing carbon loss from prescribed burning. An experiment testing different rotations suggested cessation of burning was ‘likely to be an acceptable management in the interests of conservation’ on deep peat (Rawes & Hobbs 1979), but their results did not conclusively support this (Harris et al. 2011a). The current recommendation is for burn rotations of no less than 10 years to be applied to moorlands on deep peat in England and Wales (Natural England 2007). However, the median rotation interval is closer to 20 years in England (Yallop et al. 2006) and may be as much as 50–100 years in Scotland (Hester & Sydes 1992) where current recommendations are to burn only heather taller than 20 cm (The Scottish Government 2011). There is also a move towards longer rotation intervals on blanket bog, especially where it is actively peat forming (Anon 2007). The target rotation interval dictates the proportion of moorland area a manager should aim to burn each year; for example a 10-year rotation interval requires 10% of the moor area to be burned annually. However, there is little scientific evidence to support such management strategies. Any rotation has an associated annual carbon loss, which depends on the fuel load accumulated between burns. There is, therefore, a need to quantify the carbon balance of moorland subject to prescribed burning.

This study assesses the impact of varying the prescribed-burning rotation interval on above-ground fuel load and released carbon. Fuel-load accumulation data from a chronosequence study (Harris et al. 2011a) and combustion completeness data from prescribed-burning experiments (Harris et al. 2011b) were used with measured carbon concentrations of moorland fuel-load fractions to parameterize an age-structured matrix model of above-ground fuel-load dynamics on Calluna-dominated moorland. Long-term carbon losses from above-ground fuel load were then estimated based on the stable age structure of the vegetation under various rotation intervals. Wildfires with greater associated carbon losses were included in these long-term predictions, and management strategies to minimize carbon losses under increasing wildfire return intervals are suggested. To consider the effects of delayed regeneration (vegetative or seed) on long-term carbon dynamics, a lag period was also incorporated in these models. No estimates of peat accumulation or loss are included and estimates are restricted to above-ground fuel load (comprising Calluna biomass and litter necromass). Furthermore, the effects of prescribed burning are based on data from a single site and may, therefore, have only limited general applicability, but nevertheless provide a useful first step in estimating the effect of fire on moorland carbon.

Materials and methods

All statistical analyses and model coding were performed in the R statistical environment (R Development Core Team 2009). Model code is presented in Appendix S1 in Supporting Information.

Field data collection

A chronosequence study was carried out on five moorlands within the Peak District National Park, UK, from 2006 to 2008 (Harris et al. 2011a); only the data for Howden Moor are used in this study. Here, a range of patches previously subjected to prescribed burning by the land managers were selected using an age-stratified random sampling procedure. The patches were cross-referenced with management maps, providing burned patches of known age (years since last burning) between 2 and 18 years; older patches were also identified which had not been burned for at least 35 years. Quadrat locations were selected randomly within patches. All above-ground material was harvested within a 0·25 m2 quadrat, sorted into fractions (dwarf shrubs, litter, graminoids, bryophytes and animal excrement), dried at 80 °C and weighed. Randomly selected Calluna stems were also collected and aged to verify elapsed times since burning. Full details of the sampling procedure are given in Harris et al. (2011a). The vegetation on these moorlands is extremely species-poor due to a combination of past aerial pollution (especially for bryophytes; Tallis 1998), overgrazing and wildfire (Anderson & Yalden 1981). Accordingly, the above-ground fuel load was comprised almost entirely of Calluna and litter; other fractions were present, but made up a trivial amount (~1%) of the total mass (Harris et al. 2011b), and therefore, only Calluna and litter are considered hereafter (i.e. omitting the negligible mass of graminoids, bryophytes and animal excrement at this site).

Combustion completeness (CC;% above-ground dry fuel-load reduction) was derived from 31 prescribed fires on Howden Moor (Latitude 53°28′N, Longitude 1°42′W) between March and April from 2007 to 2009. The fires were carried out as part of normal moorland management using pressurized fuel-assisted head fires (Harris et al. 2011a); the vegetation was species-poor (M19/M20) with a peat depth >50 cm. The fires were on an undulating plateau; aspect, slope and elevation were measured, but none of these variables were selected as significant in relation to fire temperatures (Harris et al. 2011a). Vegetation was harvested before burning from 0·25 m2 quadrats within the area to be burned (Harris et al. 2011b). After burning, vegetation was harvested from a secondary quadrat, 0·5 m from the first in a random direction. Pre- and postburn samples were separated into fractions, dried and weighed as before (detailed in Harris et al. (2011b). All vegetation samples were ground to pass through a 1 mm mesh, and total elemental carbon concentrations (%) were measured in randomly selected subsamples of Calluna and litter, before (n = 27) and after (n = 10) burning using a Carlo Erba Instruments NC2500 elemental analyser.

The model

The model comprised four phases: Phase 1, the long-term stable age structure of moorland vegetation was established under varying prescribed-burning rotations, and with a varying lag in regeneration; Phase 2, the amount of above-ground fuel load (and carbon therein) associated with each rotation interval in the long term was predicted; Phase 3, carbon released annually by prescribed burning under each rotation interval was estimated; Phase 4, long-term estimates of carbon lost in prescribed fires were produced and a wildfire term included to assess the impact of wildfire frequency, in combination with prescribed-burning regimes.

Phase 1: Stable age structure

Rotational prescribed burning was characterized using a Markov Chain or Leslie matrix (Leslie 1945), which is commonly used to model changes in populations through time (e.g. Bucharová, Münzbergová & Tájek 2010; Hunter et al. 2010). Over time, such models tend towards a stable age distribution, which can be used to make predictions about the population in the long term. The population is divided into groups based on age classes, in this case, elapsed time since last burning (age; years). At any point in time, the population is represented by a vector with an element for each age class indicating the size of the population currently in that class. Here, the population vector comprised proportions summing to one, to represent the proportion of total moor area in each age class. The Leslie Matrix has the same number of rows and columns as the population vector has elements, and the (i,j)th element in the matrix indicates the probability of transition from age class j to age class i at each time step. Here, probabilities in each column summed to one, as there was no change in absolute numbers, only transition between states. By multiplying the population vector by the Leslie Matrix, the population vector (or age structure) after the time step is determined. However, the long-term stable age structure can be determined directly from the Leslie matrix as the dominant eigenvector. Here, one time step represents 1 year, and the stable age structure represents the proportion of moor area in each age class in the long term under each prescribed-burning rotation.

Burning was represented in the model by transition from the age class at time of burning to age 1, representing removal of living, above-ground Calluna. The proportion of total moor area burned in each time step, annual area burned (AAB), was calculated as 1/rotation interval. To achieve this, all areas 8 years and older were burned with probability 1/(rotation interval – years not burned); for example, for a 12-year rotation, 20% (100 × 1/(12−7)) of all areas 8 years and older were burned each year to give a mean fire return time across all areas of 12 years. For the remainder of the moor area (1-AAB), vegetation in each age class aged by 1 year at the time step.

The first age at which vegetation was potentially subject to prescribed burning was 8 years. Therefore, age classes 1–7 had a probability of 1·0 of moving to the next age class. All age classes ≥8 years had an equal likelihood of being burned. Rotation intervals from 8 to 50 years were modelled. Beyond 50 years, unburned Calluna did not age further, but remained in this final age class. The model assumes no further above-ground fuel-load accumulation after 50 years because this is the age range over which field data were available. As the vegetation ages, an asymptotic relationship would be expected along with increased transfers to the peat substrate. Consideration of these aspects requires longer-term data than was available here.

To consider the effects of regeneration lag in older Calluna, the Leslie matrix was modified to include ‘negative age classes’. Calluna burned above a specified age was returned to a negative age class, where it progressed by one age class per year (with probability = 1·0), until it reached age 1 where behaviour continued as above (see Appendix S2 for example Leslie matrices). The lag period was specified to commence at age class 15 or 20 with a 5 or 10 year delay in stand establishment. Inclusion of a lag period alters overall rotation intervals, particularly in longer rotations (Table 1). Actual mean rotation interval can be calculated under the stable age structure as 1/proportion of total moor area burned annually or 1/∑(age structure × proportion burned in each age class annually), where age structure is a vector of proportions in each age class. Figures are plotted using adjusted rotation intervals up to 50 years.

Table 1. Mean rotation intervals for modelled regeneration lag periods. Inclusion of a lag period increases overall rotation intervals within the model compared with those specified, particularly at longer rotations. Actual rotation interval was calculated as 1/proportion of total moor area burned annually under the stable age structure
Specified rotation intervalActual rotation interval
No lag Lag 5 Lag 5 Age 20 Age 15 Lag 10 Age 20 Lag 10 Age 15
888888
101010·010·210·110·4
202021·822·623·525·3
303032·833·535·637·0
505053·754·157·458·3

Phase 2: Above-ground fuel-load and carbon accumulation

Once the stable age structure was known for a given rotation interval, the associated, long-term mass of above-ground, moorland fuel load was calculated. Using data from the chronosequence, linear models of Calluna and litter fuel load in relation to elapsed time since burning (on a loge–loge scale) were fitted. These were used to back-predict values and confidence limits of fuel load for vegetation aged 1–50 years for both Calluna and litter. The fuel load accumulated in i years since burning represents an estimate of fuel load in age class i of the stable age structure; where the model included a lag period, negative age classes had no associated mass. A bootstrapping procedure was then applied, whereby the proportion of moorland in each age class of the stable age structure was multiplied by a random draw from the predicted distribution of fuel load for both fractions. This was repeated 10 000 times to give the mean and 95% confidence limits for above-ground fuel load in each age class. The sum of both fractions across all age classes gave the total predicted, long-term amount of above-ground fuel load (t ha−1) for the given rotation interval and was repeated for rotation intervals of 8–50 years.

To calculate the predicted, long-term mass of above-ground carbon per hectare of moorland (Cmass; t ha−1) under each interval, the bootstrapping procedure was repeated with an additional step: each randomly drawn predicted fuel-load value was multiplied by a random draw from the distribution of measured carbon concentrations for either Calluna or litter as appropriate. Means, confidence limits and Cmass were calculated as above.

This model considers only the areas of moorland subject to burning. Areas that managers would not burn, such as sensitive or rocky areas (Anon 2007), accumulate fuel load unchecked and as such would have maximum fuel load in most cases. These are not included in total above-ground fuel load or carbon estimates, and ‘total moor area’ refers to areas with potential for management only. Only carbon in above-ground fuel load is considered here; the underlying peat is not accounted for. However, charcoal produced by burning was included within postburn litter.

Phase 3: Carbon released by prescribed burning

Annual carbon loss through prescribed burning (ClossPBA) at each rotation interval was estimated as the product of AAB and a random draw from the Cmass distribution, both for the given rotation interval, and a random draw from the CC distribution calculated from prescribed fires.

Phase 4: Long-term predictions of the impact of prescribed burning and wildfire

Carbon released through prescribed burning over the long term was calculated by running the model over 200 years (ClossPB200). To assess the impact of wildfire frequency in combination with prescribed-burning regimes on long-term carbon loss, the model was run from postwildfire conditions, that is, all vegetation in age class 1, for the given time period between wildfires (50, 100 or 200 years). By multiplying this population vector by the Leslie Matrix iteratively, the age structure at subsequent time steps was determined and used in place of stable age structure in the above calculations. Carbon lost in prescribed fires was summed over the time period between wildfires for each rotation interval and added to the predicted value of total carbon mass per hectare (Cmass). Use of Cmass to represent carbon loss in a wildfire assumed maximum fuel-load consumption, that is, that CC was 100% of all age classes (including <8 years). The 50- and 100-year values were then multiplied to give values for a 200-year period. Where a lag in regeneration was modelled for prescribed burns, all vegetation was subject to the same lag period following a wildfire, regardless of age before the wildfire, to simulate the increased severity of wildfire compared with prescribed fire. Rotation interval here refers to prescribed burns only and does not incorporate the effect of a wildfire on overall fire return interval, which will clearly be reduced.

Results

The chronosequence study revealed a linear, loge–loge relationship between vegetation age (years) and above-ground fuel load (Fig. 1; Calluna: r2 = 0·89, < 0·0001; Litter: r2 = 0·90, < 0·0001). Combustion completeness derived from prescribed fires on Howden Moor was 71·4 ± 2·6% for Calluna and 54·5 ± 2·8% for litter (mean ± SE; Fig. 2). The variation in CC values measured here reflected a variation in fuel moisture content of 42–194%. Carbon concentration was calculated as 48·3 ± 0·1% for Calluna and 49·0 ± 0·1% for litter. For details of species cover, above-ground fuel load and combustion variables see Harris et al. (2011a,b).

Figure 1.

Above-ground fuel load (t ha−1) of Calluna (top) and litter (bottom) accumulated following prescribed burning on Howden Moor, Derbyshire. The relationship is linear on a loge–loge scale (left). Dashed lines are 95% confidence limits. Equations are Calluna: loge(mass) = 1·2loge(elapsed time +1) −0·9) and litter: loge(mass)=1·1loge(elapsed time +1) −0·7. Raw data are shown (right).

Figure 2.

Total above-ground fuel load (t ha−1) before (light) and after (dark) prescribed burning; 31 fires ranked by preburn fuel load. Mean values ± SE are reported (n = 4).

Stable age structure

The stable age structure of vegetation varied depending on the prescribed-burning rotation interval (Fig. 3). Age classes were gathered into the Calluna growth phases first described by Watt (1947), (Seedling = 1, Pioneer = 2–8, Building = 9–15, Mature = 16–19, Degenerate = 20+ years). The ages of the early stages are reasonable for the study site, but the mature and degenerate stages merge as vegetation tends to regenerate by layering (MacDonald et al. 1995). However, we present this to be consistent with other published data and because these morphological stages are often used by land managers to assess the quality of Calluna vegetation. Unsurprisingly, at shorter rotation intervals when burning occurred more frequently, the population tended towards a younger age structure dominated by pioneer Calluna, whereas at longer intervals, the population comprised mostly degenerate Calluna.

Figure 3.

Long-term, stable age structure of Calluna vulgaris-dominated moorland vegetation under fixed prescribed-burning rotation intervals. Proportion of population in each age class (left) and growth phase (right) for 8-, 15-, 25- and 50-year rotation intervals.

Fuel load and carbon released in fires

The predicted, long-term, above-ground fuel load (and associated Cmass) increased with rotation interval, that is, when burning was less frequent, the resulting fuel load was greater (Fig. 4, Table 2).

Table 2. Mean predicted total (Calluna + litter), long-term, above-ground fuel load and carbon mass (t ha−1) across the entire moorland under the stable age structure and various prescribed burning rotation intervals
Rotation intervalTotal fuel loadTotal Cmass
106·72·1
2523·614·7
5041·620·2
Figure 4.

Predicted long-term, above-ground fuel load and carbon mass (t ha−1) of Calluna (a) and litter (b) averaged across the moorland under various prescribed-burning rotation intervals (with no regeneration lag). Means (continuous) and 95% confidence limits (dashed) from 10 000 bootstrapped values are shown.

Carbon lost annually in prescribed fires (ClossPBA) was minimized at longer rotation intervals (Fig. 5), although where regeneration lag was absent, short rotations also offered reduced carbon loss compared with intermediate (15–25-year) rotations. However, long-term calculations including wildfire showed a different pattern (Fig. 6); in the absence of lag, more regular wildfires (50 and 100-year return intervals) resulted in carbon loss minima at short prescribed-burning rotations, Furthermore, at a 200-year wildfire return interval, carbon loss was minimized at both short and long rotations.

Figure 5.

Modelled annual carbon loss (t ha−1) due to prescribed burns (ClossPBA) under varying rotation intervals. Regrowth after burning was modelled as either immediate in all age classes (no lag) or subject to a 5- or 10-year regeneration lag in age classes above 15 or 20 years.

Figure 6.

Modelled carbon loss (t ha−1) over 200 years with respect to prescribed-burning rotation interval at 50-year (solid line), 100-year (dashed line) and 200-year (dotted line) wildfire return intervals (diamonds indicate minima). Regrowth after prescribed burning was either immediate irrespective of age burned (no lag) or subject to a 5- or 10-year time-lag when burned at age 15 or 20 years upwards. Eight-year rotation intervals have been omitted to aid interpretation, as model specification dictates that all vegetation is burned every 8th year causing large deviations between burn and nonburn years in postwildfire conditions.

In most cases, carbon loss (ClossPB200) minima occurred at 9-year rotation intervals regardless of wildfire return interval or lag. With a 200-year wildfire return interval, carbon loss was minimized either at a 9-year or 50-year prescribed-burning rotation, and was greatest at intermediate rotations, in all but the most severe lag treatment (10-year lag periods acting on vegetation aged from 15 years), which resulted in carbon loss minima between 15 and 20 years.

Discussion

Moorland above-ground fuel load in the UK is most commonly manipulated using prescribed burning. Cutting is also possible (Miller & Miles 1970) although more expensive, and the use of machinery can damage fragile peat soils. Moreover, wildfire risk is only reduced if cut vegetation is removed from the site, and carbon emissions only prevented if it is not subsequently burned or decomposed. Grazing is also used, but if pressure is too high, species composition can move from dwarf-shrub- to graminoid-dominated vegetation, changing its conservation value (Anderson & Yalden 1981). Moorland management has changed in recent years with reduced grazing pressure (DEFRA 2010), usually as a result of agri-environment schemes (MAFF 1996). Prescribed burning may, therefore, be the most cost-effective tool for managing moorland ecosystems that contain fire-adapted species (Måren et al. 2010). In most areas, combinations of prescribed burning and light grazing are enough to prevent ingress of trees, although on Howden Moor, some (Betula spp., Pinus sylvestris) are colonizing the fringes. However, prescribed burning is not without drawbacks, as it impinges on carbon accounting and may increase soil nutrient concentrations and coloration (DOC) in water draining from moorlands (Yallop & Clutterbuck 2009), which increase costs of purification for human consumption (Worrall et al. 2010). As noted above, prescribed burning of moorland vegetation on peat is a cultural management practice particularly prevalent in Great Britain. Thus, our results apply at present to a very limited subset of current vegetation types within the boreal region. However, if the boreal regions become warmer and drier in summer, wildfire risk will increase and prescribed burning may be one technique used to minimize damage (Albertson et al. 2010). If this occurs, then the approach used here may be used to inform future management strategies for dwarf-shrub communities at a much larger international scale. Wildfire events have already been shown to cause substantive damage in the boreal region (e.g. Kasischke & French 1995). Under such circumstances, the ability to assess the relative carbon losses of prescribed fires versus wildfire will be essential for carbon accounting. We see this paper as a first step towards this for dwarf-shrub vegetation.

Carbon accumulation at a case-study site was modelled based on chronosequence vegetation data (Harris et al. 2011a) and measurements of carbon loss during prescribed burning (Harris et al. 2011b). The impact of delayed postburn regeneration was also assessed. Increased prescribed-burning frequency shifted the age structure of the moorland to a younger state, which inevitably reduced overall fuel load and increased carbon loss. However, when a wildfire was included in the model, three important conclusions were drawn:

  1. There was clear interaction between prescribed-burning rotation interval and wildfire return interval. At 50- and 100-year wildfire return intervals, carbon losses were minimized by short prescribed-burning rotations.
  2. At a 200-year wildfire return interval, prescribed burning had less effect on carbon loss, which was highest at intermediate rotations in all cases except the most severe lag specification (10-year lag period acting from age 15 years), whereas intermediate rotations minimized carbon loss.
  3. A lag in vegetation recovery, both in terms of the age that lag took effect and lag duration, also shifted the minima slightly and resulted in less carbon loss (due to less vegetation across the entire moor at any one time).

Thus, appropriate management will depend on the wildfire return interval, which is unpredictable, particularly as ignition often results from human involvement in the UK (accidental or arson; Albertson et al. 2009). However, our modelled values straddle the approximate natural background fire incidence of 125 years between 1000 and 3000 years ago, which were derived from peat cores on nearby Robinson's Moss (K. Halsall, unpublished data). As the probability of wildfire incidence increases, prescribed burning may reduce carbon losses at short rotation intervals. For example, under a 50-year wildfire return interval, an 8-year prescribed-burning rotation would result in 34% less carbon loss than a 50-year rotation (75 and 114 t ha−1, respectively, over 200 years with no modelled lag). Without knowledge of the existing wildfire return interval, it is difficult to suggest where recommendations should currently lie, but if wildfire incidence increases in future, it is likely that more frequent prescribed burning will mitigate carbon loss. These results suggest that if wildfire return interval is 100 years or less, prescribed-burning rotation intervals between 8 and 18 years could minimize carbon loss. However, this does not account for either reduced wildfire incidence due to fuel-load reduction or firebreak creation by prescribed burning, or increased risk of wildfire due to escaped management fires, both of which would alter the optimum rotation interval.

A no-burn policy is also an option, particularly when wildfire is less frequent. However, fuel load would be at maximum levels and, when wildfire did occur, could produce an intense and damaging fire, with slow ecosystem recovery (Maltby, Legg & Proctor 1990) and a potential maximum carbon loss of ~42 t ha−1 per wildfire event on Howden Moor (assuming 100% CC and no peat ignition, Table 2). Wildfires can burn for many days covering hundreds of hectares, for example at Angelzarke Moor, Lancashire, UK, a wildfire covered 10 km2 in spring 2012 (McMorrow & Cavan 2011 2011), and often consume large proportions of above-ground vegetation, some surface peat and potentially subsurface peat (Maltby, Legg & Proctor 1990). Inevitably, any fire will release carbon stored in vegetation to the atmosphere as CO2 and smoke, and into soils as charcoal, particulates and possibly DOC. However, prescribed fires and wildfires exert different pressures on the landscape, with wildfires in some instances resulting in greater fire intensities (rate of energy released), fire severities (consumption of organic matter) and ecosystem response (alteration of functional processes; all sensu Keeley 2009). Available fuel and its structure are key factors in determining fire characteristics; with greater fuel-load availability, there will be increasing fire intensity and severity, associated with greater combustion completeness (Keeley 2009).

At the rotation intervals recommended here (8–18 years), Calluna should recover at least partly by resprouting (e.g. Miller & Miles 1970; Davies et al. 2010). Calluna can produce new shoots within a year of burning and regain dominance after 2–3 years where vegetative propagation is present (Vandvik et al. 2005) or 5–7 years where regeneration is from seed. Furthermore, regeneration characteristics of young heather can be re-established after only two rotations following restoration from degenerate condition (Velle, Nilsen & Vandvik 2012). While conditions are different at the site considered here to those in Velle's study, the vegetation is productive with active layering and is likely to show rapid regeneration. Therefore, if older Calluna is burned as part of a wildfire risk-reduction strategy, it is likely to resume dynamics characteristic of managed moorlands rapidly (<20 years based on recommendations herein). Not only does degenerate moorland constitute a greater carbon source in the event of a wildfire, the increased fuel load may also result in fires that are more heterogeneous in behaviour (Davies et al. 2010) and more difficult to control, thus endangering life and property. Old stands also have reduced species diversity (Hobbs & Gimingham 1984b; Harris et al. 2011a) and greater moss and litter cover, which is rarely fully consumed by prescribed fire and slows Calluna regeneration (Davies et al. 2010). However, older stands may provide refuges for species not found elsewhere (Davies & Legg 2008).

The models presented here may somewhat misrepresent current burning practice, in that the land area burned annually is divided proportionally across age classes. In reality, managers may focus on burning Calluna in the building stage to maximize regeneration, or older heather may be burned to reduce fuel load (Davies et al. 2008, 2010). We suggest that a shift to burning uniformly across age classes would benefit carbon and fuel management by reducing fuel load and maintaining fire breaks. In the long term, this approach would maintain the stable age structure and, if applied in a mosaic pattern, would maintain species and structural diversity for the full range of moorland species. Furthermore, given that delayed regeneration following fire will delay the resumption of carbon storage, it might be expected that more regular prescribed burning, producing younger vegetation overall, would be preferred to reduce the proportion of vegetation subject to delayed regeneration.

Leslie models can predict the development of a successional sequence and have been shown to tally with field data on successional stages in some systems (Usher 1981), including postfire heathlands (e.g. Hobbs 1983). As the system considered here is a single vegetation type with one dominant species, and transitions are between age classes not successional stages or dominant species, the Markov model should give a reasonable approximation to their dynamics. Models are simplifications of reality and can generate unrealistic situations. For example, here, an 8-year rotation interval applied postwildfire leads to the entire moor being burned every 8th year with no burning in between, which would be highly impractical and unlikely to be adopted. However, models provide an insight into patterns and processes and the potential scale of carbon loss. The fuel-load and carbon loss estimates are an approximation, but the derived trends due to lag and wildfire frequency are informative for preparing management plans for a warming climate and can be considered alongside more complex species and ecosystem model predictions (e.g. van Tongeren & Prentice 1986; Smith, Prentice & Sykes 2001). Further information is needed on the response of Calluna growth phases to wildfire, with which the model could include more realistic estimates of wildfire-induced CC and the impact of different fuel ages. This would allow more detailed management planning and consideration of which areas should be subject to prescribed burning for wildfire prevention. Furthermore, CC data from a range of fire locations would allow the model to be generalized to other moors.

This study is confined to above-ground fuel load; no attempt has been made to accommodate annual carbon fluxes into the below-ground system (plant roots, peat). Similarly, peat combustion is not considered, thus carbon loss during wildfire will be significantly underestimated in conditions where peat is ignited. However, inclusion of charcoal in the postburn fuel-load estimates may lead to over-estimates of carbon loss in managed and wildfire. The burns used to determine CC were routine management burns on relatively young vegetation on a moor managed for grouse production. The moor is also very productive and is at the drier end of UK moorland vegetation (Lee et al. 2013). As such, the CC distribution may not well represent burns on older vegetation or wetter sites. However, data from prescribed burns in Northumberland moors (K.A. Allen, unpublished data) suggest CC of very old, degenerate Calluna is ~62%. The value from Howden Moor was used as it is representative of burns at the site modelled and on vegetation including younger age classes. The CC values used here were within the literature range of CC values for Calluna (94% for autumn burns and 55% for spring burns: Kayll 1966) and total fuel load (83%: Hobbs & Gimingham 1984a; 79%: Legg, Davies & Gray 2010). Litter values are harder to compare due to differences in measurement techniques, although here they may be somewhat higher due to the high productivity and relatively dry nature of the site. The use of 100% loss during wildfire is a compromise, and assumes complete loss of above-ground vegetation, which is unlikely in a fire of low- to moderate-severity and hence is an over-estimate of losses. It does, however, represent the maximum loss of above-ground fuel load and carbon in severe fires where the fire would extend into the peat resulting in an under-estimate. Alternatively, a percentage loss drawn from the distribution of combustion efficiencies might be used. However, the range was very large, and it is likely that this would under-estimate losses. Clearly, better data on the impact of wildfires of different severities are needed for British moorlands to progress this further. Additionally, no account is taken of the effect of managed fire on wildfire ignition risk or area of spread, and the results do not apply to lowland heath or upland Calluna-dominated communities on mineral soils. This precludes a complete understanding of ecosystem processes, but the fuel-load estimates herein allow initial comparative assessment of change in the above-ground carbon pool under a range of wildfire scenarios. Our results, therefore, provide information to guide policies for the future sustainable management of European heaths and moors.

This study suggests that more regular prescribed burning may minimize carbon loss from the combustion of above-ground fuel load, particularly when wildfire is more frequent. We also recommend that burning is applied across the range of Calluna growth phases, to reduce fuel load and increase biodiversity in degenerate moorland (Harris et al. 2011a; Lee et al. 2013). Areas excluded from burning are not considered in this study, but may represent maximum fuel load as they accumulate fuel load towards a potential asymptote with further net production balanced by incorporation into the peat. As such, their location should be considered carefully, and effective fire breaks positioned to protect large areas. This work increases understanding of the role of prescribed fire in managing wildfire and carbon budgeting. However, more detailed understanding of individual areas is needed to refine these guidelines into targeted burning plans (Davies et al. 2008; Velle, Nilsen & Vandvik 2012). Optimizing the balance between moorland ecosystem services will require careful consideration of which areas should be burned to maximize regeneration.

Acknowledgements

We thank Geoff Eyre, Mark Osborne, Heather Trust, Moors for the Future, Moorland Association, Royal Horticultural and Botanical Society of Manchester and the Northern Counties and the BiodivERsA FIREMAN program (NERC/Defra) for funding. Sabena Blackbird carried out carbon analyses. Richard Bradshaw, Matt Davies, Colin Legg and an anonymous referee provided valuable comments on the manuscript.

Ancillary