Fuel, fire and cattle in African highlands: traditional management maintains a mosaic heathland landscape

Authors


Summary

  1. Shrubland ecosystems are often inherently flammable due to a canopy structure favourable for fire propagation. At the same time, the fuel bed is not spatially uniform, but a complex mix of shrubs and herbaceous vegetation that will change with time since fire. These patterns are further influenced by megaherbivores capable of consuming large quantities of biomass that otherwise would enter the fuel bed, but the net effects for temporal thresholds of flammability are poorly known.
  2. We quantified post-fire fuel succession and effects of free-ranging cattle in high-elevation Erica shrublands in Ethiopia where traditional fire management is still practised above the treeline at 3500 m, despite being challenged by authorities.
  3. We found a near-linear accumulation of canopy fuel at 2·6 Mg ha−1 year−1, but stands <5 years old did not burn due to spatial separation of individual Erica shrubs before canopy closure and lack of fine dead fuels.
  4. Inside cattle exclosures, Erica height growth was nearly three times faster and reached the assumed flammability threshold c. 3 years earlier than in browsed/grazed stands, where cattle also kept herbaceous vegetation between shrubs short, thus eliminating litter that could otherwise bridge the discontinuous fuel bed in early succession.
  5. Modelling of fire behaviour indicated progressively higher fire intensity and rate of spread for stands >5 years. But if stands escape fire for several decades, flammability again decreases as canopy fuels become increasingly separated from the ground. This may be the ultimate reason for sharp treelines on many tropical mountains where fire is confined mainly to higher elevations.
  6. Synthesis and applications. In shrublands where dominant plants can outgrow consumption by large herbivores, provision of good cattle habitat typically requires fire. We found that fire and cattle interact to maintain a relatively stable system, where fuel limitation in early succession creates fire breaks that prevent landscape-wide wildfires. The same negative feedback protects the Erica from degradation by too frequent fires. As shrublands can have widely different compositions, the response to variation in fire frequency and herbivore pressure is likely to differ, but for sustainable management fire and grazing have to be treated in consort.

Introduction

Fire exerts a main control on vegetation composition and structure in many shrub-dominated ecosystems (Keeley et al. 2012). Fires in this type of vegetation are often difficult to suppress due to fast spread rates and high intensities. Beside weather factors, fuel quantity and quality are important determinants of fire behaviour. After each fire event, which consumes fuel and typically completely top-kills the shrub-layer, there will be a long period of dramatic changes in both fuel quantity and quality, causing gradual changes in potential fire hazard over a number of years (Rothermel & Philpot 1973). An important effect is that this can define the shortest possible fire interval (Schimmel & Granström 1997). Once the minimum fuel requirements for fire propagation are met, further changes will affect the potential rate of spread and fire intensity (Fernandes, Catchpole & Rego 2000).

Prescribed burning is currently discussed as a management tool to reduce the increasing fire hazards in fire-prone shrubland ecosystems, based on the assumption that fuel elimination will create fire breaks in the landscape (Minnich & Chou 1997). This is sometimes disputed on the grounds that the fire break effect may last only a short period (Moritz et al. 2004) or that older stands harbour fire-sensitive species and therefore should be conserved for biodiversity reasons (Haslem et al. 2011). Frequent burning may also favour fire-adapted invasive species (Keeley 2006; van Wilgen et al. 2010). It may seem like a paradox that while fire now is re-introduced in developed parts of the world (Fuhlendorf et al. 2009), fire use is still questioned in poorer countries where traditional fire management prevails. Authorities often perceive fire as destructive to ecosystems and aim to control it (Laris & Wardell 2006; Angassa & Oba 2008), even though it can be a prerequisite for pastoralist livelihoods (Johansson et al. 2012).

Fuel succession, the development of fuel beds over time, is primarily controlled by rates of species replacement, plant growth, litter production and decomposition rates (Riggan et al. 1988). All these processes may further be influenced by herbivores consuming selected parts of the vegetation, altering species composition, decreasing total fuel load, changing the relations between different fuel fractions and increasing decomposition rates. Effects of megaherbivores on shrubland fuels vary from insignificant (Williams et al. 2006) to large (Davies et al. 2010; Raffaele et al. 2011), and the response should depend on herbivore pressure as well as vegetation composition. Naturally, herbivores impact palatable vegetation most, which at the same time may be the most easily decomposed (Cornelissen et al. 1999) and therefore also least important as fuel. It can be assumed that in shrublands, herbivore effects will be greatest in the early post-fire years, when pasture quality is highest (Fuhlendorf et al. 2009). Consumption of potential fuel should also alter temporal fire thresholds (Twidwell et al. 2013), for example by slowing down fuel build-up, increasing minimum fire-return intervals and decreasing fire intensity.

Despite the large interest in fuel reduction in shrub-dominated ecosystems, there are few studies detailing post-fire successional changes in fuel under the influence of megaherbivores and concomitant effects on potential fire behaviour (Keeley et al. 2012). Here we use a chronosequence of stands in the East African ericaceous zone to analyse changes in fuel quantity and quality over time-since-fire. A set of permanent exclosures allowed us to estimate the effect of cattle grazing on fuel-accumulation rates. To analyse how this translates into fire behaviour, fuel and weather data were fed into the BehavePlus model (Heinsch & Andrews 2010). In addition, field observations of fire behaviour in relation to fuels were collected during four dry seasons. Our aim was to answer the following questions, which are central to management policy: (i) How does the fuel bed develop over time-since-fire and how is this affected by cattle grazing and browsing? (ii) How do changes in fuel load with time-since-fire translate into fire behaviour in complex shrub-dominated ecosystems? (iii) What are the implications of these relations for sustainable land management?

Materials and methods

Study Area

The study was undertaken in Bale Mountains in the Southern highlands of Ethiopia within an area of about 33 × 11 km, centred at 06° 50′ N and 39° 15′ E, at elevations between 3400 and 3700 m. The bedrock originates from Tertiary lava and soils are dark silty loams, with acidic topsoils rich in organic matter (Yimer, Ledin & Abdelkadir 2006). The climate is cold and wet, except in the short dry season, normally lasting December–January, or into March–April during drought years (Miehe & Miehe 1994). Mean annual precipitation is c. 1740 mm (pers. obs.). Small amounts of rain fall also in the dry season. There is a large diurnal variation in temperature, with average daily max and min temperatures c. 15 and 5 °C (pers. obs.). Temperature fluctuations are larger in the dry season, when night frost is common. The vegetation is dominated by two potentially tree-forming species, Erica arborea (L.) and Erica trimera [(Engl.) Beentje]. At high elevation, the Erica is burnt on short rotation by pastoralists (Johansson et al. 2012) in order to improve pasture for free-ranging cattle Bos taurus indicus. At about 3500 m elevation, there is a distinct treeline, below which E. trimera, the taller of the two species, forms a closed-canopy, 8 to 11 m-tall, forest (Fetene et al. 2006), which is never or only very rarely burnt. The heathland vegetation above the treeline forms a continuous vegetation belt with a mosaic of regenerating stands of different age (time-since-fire), which rarely become taller than 2 m before being burnt again. After each fire, both Erica species re-sprout vigorously from buried lignotubers and young stages are dominated by discrete clusters of new green Erica shoots, with herbaceous vegetation in-between that is kept very short due to heavy grazing pressure. In older stages, the Erica canopy cover is more or less continuous, successively increasing in height (cf. Fig. 1).

Figure 1.

Erica stands with increasing time-since-fire. (a) 2 years post-fire, inside a grazing exclosure. Note the remaining dead stumps and the poor grass growth despite cattle exclusion. (b) 5 years post-fire, with almost closed canopy. (c) 11 years post-fire with a high, and vertically continuous fuel load including a high proportion of persistent fine dead twigs. (d) c. 50 years since fire (a stand situated just below the treeline) with the canopy fuels vertically separated from the surface fuels.

Chronosequence Fuel Quantification

We quantified Erica fuel by destructive sampling in a chronosequence of stands of different time-since-fire (1–22 years). Stands (i.e. patches of uniform-aged shrubs regenerating after fire) differed in size from c. 0·5 to 2·5 ha. Both Erica species form annual rings (Johansson et al. 2012), and stand age was determined by ring counts on three individuals per species per stand. All biomass samples were collected in the dry season, January–February, 2007 and 2008. Shrub fuel was destructively sampled on representative plots within the Erica canopy (n = 54). Plot sizes were 0·2–1 m2, depending on shrub size. Shrub canopy height was measured at five points across the plot before harvest. We categorized the plot as ‘browsed’ if any browsed shoot tips were observed, or ‘unbrowsed’ if there were no signs of browsing, neither on the shoot tips, nor lower down the stems. (It is possible to trace browsing 1–4 years back on the shoots due to the presence of a kink at the browsing point where the side bud took over as the dominant stem.)

All biomass in the shrub canopy and litter/moss layer was harvested. Live and dead Erica stems were cut at the soil surface and further sectioned into different biomass fractions (Table S1, Supporting Information). In young stands, we sorted all biomass, in older stands representative sub-samples only. To estimate moisture contents and dry weights, each fraction was put into separate plastic bags, sealed and weighed on a portable scale, transported to a local lab, dried at 80 °C to constant weight and re-weighed to nearest 0·1 g. All moisture contents are expressed per dry mass.

Mass of each biomass fraction was first calculated per unit shrub area for each plot. To convert this into biomass per unit total area (including canopy gaps), we used a large data set from the same area, of Erica canopy cover vs. age. This was recorded in 330 (10-m long) line transects (cf. Bauer 1943) spanning a range of stand age. In each transect, we noted the proportion of the line covered by the vertical projection of shrubs, soil, grass/herb or moss and measured the dominant heights of the two Erica species, the grass/herb vegetation and moss (mainly Breutelia borbonica). A Lowess regression (locally weighted scatter-plot smoothing) was applied to the cover data over stand age. Fuel data per unit shrub area, from the destructively harvested plots, were then multiplied with the average shrub cover for the respective stand age (from the Lowess regression) to obtain functions of total biomass (including gaps in-between shrubs) vs. height and age expressed as Mg ha−1.

Cattle Exclosures

In 2005 and 2006, we built three cattle exclosures (10 × 10 m) in recently burnt heathland stands at 3510–3630 m elevation. The exclosures were too small to allow repeated destructive sampling. Instead, an estimation of Erica fuel mass inside and outside exclosures was obtained by using the function of fuel mass vs. height that was established for the destructively sampled stands (r2 = 0·90) and adjusting for Erica canopy cover obtained from 10-m-long permanent line transects (n = 5) inside and outside exclosures. In those, height and canopy cover was measured (as above) yearly until January 2011. To estimate grass/herb biomass inside and outside exclosures, we used permanent grass/herb vegetation plots (60 × 60 cm, n = 7) positioned between Erica shrubs. In these, cover of different grass/herb species and vegetation height was estimated yearly between 2006 and 2008. To create a function of grass/herb biomass vs. cover and height, we destructively sampled grass/herb vegetation in plots (30 × 30 cm) inside and outside exclosures (n = 12) covering a range in vegetation height. The obtained function (r2 = 0·91) was then used to estimate an average grass/herb biomass using the vegetation data (cover and height) from the permanent plots.

Erica trimera Forest

To quantify surface fuels with and without grazing in old E. trimera forest, we built two 6 × 6 m exclosures, in closed-canopy forests at 3440 and 3400 m just below the treeline. The forests were c. 11 m tall, with a canopy base height of c. 8 m and a basal area of 30 and 38 m2 ha−1, respectively. Ring counts on felled Erica stems indicated stand ages of c. 90 years. One year after fencing, we destructively sampled the field layer in 30 × 30 cm plots, inside and outside each exclosure (n = 4).

Weather Data, Fire Observations and Fire-Behaviour Modelling

Temperature and relative air humidity (RH) was recorded by radiation-shielded data loggers placed at two locations (at c. 3450 m) in the study area, running for 2 years (2006–2007). Rain data were collected by local staff at the same sites during 4 years (2006–2009). At all times when working in the heathland zone, we recorded all observed fires. For each fire (which, depending on vantage point, could sometimes be kilometres away), we recorded wind speed, temperature and RH using hand-held instruments, slope and vegetation height (if possible to estimate), and fire-behaviour indicators such as smoke colour, flame length and rate of spread (when possible to estimate).

To estimate height, cover and age of stands that were subject to burning over the course of 1 year, we made landscape-stand transects, in total 5·8 km long. The visible borders between stands of different time-since-fire were positioned using GPS. We aged the stands by annual ring counts at the base of Erica stems and measured the heights of five shrubs. Area cover of Erica spp. was visually estimated. The same landscape transects were measured twice, in February 2007 and in February 2008, which gave us information on the pre-burn height and age for the stands that had burnt during this period.

Modelling of fire behaviour (rate of spread, flame length and fire-line intensity) in relation to stand age was done using the BehavePlus model version 5·0 (Heinsch & Andrews 2010) applied to the chronosequence fuel data and typical scenarios of fuel moisture and wind. For details on the modelling, see Appendix S1 in Supporting information.

Statistical Analyses

To describe relationships between stand age or height and quantities of the different biomass fractions, we used linear and non-linear regressions. To test whether the quantity of Old-generation stems had any browsing protection effect, we used linear regression of Old-generation mass on net annual biomass accumulation (total biomass/age). To test the effect of Erica species and browsing category (browsed/unbrowsed) on biomass of different fuel categories, we used a two-way anova on natural-log-transformed data (to meet anova assumptions) with species and browsing category as fixed factors, and ln (age) and ln (height) set as covariates. To test the effect of cattle exclusion on Erica height and area cover in the permanent line transects, the data were checked for normality and homogeneity of variance and then analysed by anova with four factors: treatment (fenced/browsed), age, transect number and site. Transect number was nested within treatment and site and so was the interaction age × transect number. Treatment and age were fixed factors, while transect number and site were random. For the height model, an additional fixed factor, species, was included (with its interactions). For Erica height data, the unit of replication was the individual shrub, and for area cover, the unit of replication was the line transect. The estimated grass/herb biomass inside and outside exclosures was analysed by anova with treatment (fenced/browsed) as fixed factors and site as random factor, the unit of replication was the small permanent vegetation plots. All statistical analyses were performed using Minitab 16.

Results

Heathland Fuel Succession

Relationships between stand age (time-since-fire) and fuel loads of different fuel categories showed slightly different patterns in relation to time-since-fire (Fig. 2). Total fuel, 1-h fuel and 10-h fuel increased linearly with time over the whole chronosequence, whereas live woody fuel <6 mm levelled off within c. 10 years. Total fuel accumulation (i.e. all biomass excluding Live stem >6 mm) averaged 2·6 Mg ha−1 year−1, of which 96% consisted of dead or live Erica material. Total biomass accumulation (including Live stems >6 mm) was near-linear (not shown) and averaged c. 3·1 Mg ha−1 year−1. Old-generation stems were only present in quantity 1–4 years after fire (due to rapid decomposition), and biomass was highly variable between stands (0·01–23 Mg ha−1). A negative non-linear function was a reasonably good fit to the data (Fig. 2). For stands 1–4 years after fire, there was a positive relationship between Old-generation biomass and average annual shrub biomass accumulation (total biomass/age) (= 14·5, < 0·001). Further, there was a negative relationship between the proportion of the shrub canopy that had been recently browsed and Old-generation biomass (= 15·5, < 0·001).

Figure 2.

Biomass per hectare of different fuel categories in relation to time-since-fire (n = 54). Total fuel excludes Live stem > 6 mm and Old-generation stems. See Table S1 (Supporting information) for a description of the fuel categories.

According to the variance analyses of the fuel categories (Table 1), there was no effect of Erica species per se on fuel quantities. For all fuel categories, age and/or height explained much of the variation. There was an effect of recent browsing on Total fuel and LW <6 mm categories. Relationships between stand height and mass of various biomass fractions (Fig. S1, Supporting information) were similar to the relationships between age and mass, but regressions in most cases had higher r2.

Table 1. Summary of the analysis of variance (two-way anova) of species and browse category (br) and with ln (age) and ln (height) as covariates on the (ln-transformed) fuel categories in Fig. 2. Significant (P<0.05) effects and interactions in bold
 d.f.ln (Total fuel)ln (LW < 6 mma)ln (LW > 6 mmb)ln (1-h fuelc)ln (10-h fueld)
F P F P F P F P F P
  1. a

    Live woody fuel < 6 mm including leaves.

  2. b

    Live woody fuel > 6 mm diameter.

  3. c

    Dead fuel < 6 mm diameter.

  4. d

    Dead fuel 6–25 mm diameter.

ln (age)1 15·06 0·000 4·92 0·031 9·17 0·004 12·55 0·001 4·23 0·045
ln (h)1 51·47 0·000 34·00 0·000 0·570·453 7·53 0·008 5·05 0·029
Browsed1 10·51 0·002 11·59 0·001 0·960·3311·040·3130·340·561
Species10·400·5320·010·9350·090·7612·490·1210·070·796
br × ln (age)1 6·54 0·014 12·43 0·001 3·450·070    
Error45          
Total53          

Grazing Exclosures

Both Erica species were taller inside exclosures (= 526·7, = 0·028), and the difference was greatest for the dominant species E. trimera (for which heights are shown in Fig. 3a). Five years after fire, the shrub canopy was on average nearly three times taller inside than outside exclosures. In the 330 Erica transects sampled across the landscape, there was a large variation in height at any given age, spanning much of the range in height that was observed inside vs. outside exclosures (Fig. 3a). Net height increment in browsed vegetation was near linear and averaged c. 11 cm per year.

Figure 3.

(a) Height of Erica trimera in relation to time-since-fire. Symbols connected with lines represent average heights in the permanent fenced (□) and browsed (○) plots. The small dots (and the linear regression) represent data from the large data set of line transects in the grazed landscape. (b) Erica canopy cover in relation to time-since-fire. Symbols as in panel (a) but the trend line is a Lowess regression.

There was also an effect of cattle exclusion on Erica canopy cover (both species combined) (= 38·1, = 0·023) (Fig. 3b). The average proportion of E. trimera out of the total Erica cover was 60% (SE = 1·5, n = 330). On average, 50% shrub cover was reached 4 years after fire, but with a high degree of variation between individual transects, and 75% shrub cover was reached after c. 10 years. Five years after fire, the total shrub fuel mass was estimated to be ca three times higher inside exclosures than outside (Fig. 4).

Figure 4.

Estimated fuel mass inside and outside exclosures over time. Average of the three exclosures. Error bars = 1 SE.

Vegetation re-colonization was rapid and cover of the grass/herb vegetation both inside and outside exclosures peaked (at 43% on average) after 1–3 years, then decreased to c. 30% after 5 years, as the Erica cover expanded. Common species were Andropogon amethystinus, Trifolium cryptopodium, Haplocarpha rueppellii and Alchemilla abyssinica with little difference inside and outside exclosures. The proportion of grass in the grass/herb vegetation was on average 41% (SE = 5·2, n = 21). Outside exclosures, the grazed vegetation height was invariably 0·5–2 cm, during all years. Inside exclosures heights were highly variable, depending on species (cf. Fig. 1a). Two years after fire, the estimated grass/herb biomass was significantly higher (= 10·9, = 0·041) inside exclosures (173 g m−2) than outside (67 g m−2). Average grass/herb moisture content was 140%. Area cover of moss increased from 0 to c. 10% in 5 years and was present only under shrubs.

In the E. trimera forest below the treeline, the field layer was dominated by procumbent mesophyllic herbs (e.g. Alchemilla abyssinica, Geranium arabicum, Parochetus communis and Trifolium semipilosum var. brunellii) which attained c. 10 cm height and a biomass of 140 g m−2 (SE = 15·9, n = 4) inside exclosures, 1 year after fencing. Outside exclosures, the grazed grass/herb vegetation was 0·5–2 cm tall, with a biomass of 65 g m−2 (SE = 6·2, n = 4).

Fire-Behaviour Modelling

The fire-behaviour model gave zero or near zero fire-line intensity, rate of spread and flame lengths for all stands younger than 3 years, at the tested sets of weather conditions and slopes (Fig. 5). For stands older than c. 6 years, there was a dramatic increase in fire-line intensity and flame lengths with increasing time-since-fire. Terrain slope (0–40%) had little effect on fire behaviour. Increasing wind speed from 3 to 5 m s−1 doubled the intensity in stands taller than 60 cm. Reducing fine-fuel moisture content from 15% to 10% gave on average a 50% increase in intensity in stands taller than 60 cm.

Figure 5.

Modelled flame length in relation to time-since-fire for all sampled stands at normal fire weather conditions (slope = 0%, fine-fuel moisture content = 15%, wind speed = 3 m s−1).

Fire Observations and Flammability Thresholds

We observed in total 192 fires in the heathland zone during four dry seasons. Most fires occurred in the middle of the day, at a relative air humidity below 40%. Ambient wind speed was on average 3 m s−1. Flame lengths were up to c. 8 m in c. 200 cm tall vegetation (Fig. 6). No fires were observed in vegetation shorter than c. 60 cm and all observed fires terminated when reaching short vegetation. In the re-sampled landscape transects, no stands shorter than 60 cm, or younger than 5 years, had burnt during the one-year period of observation (Fig. 7).

Figure 6.

Typical fire behaviour in the dry season (photograph taken 4 February 2008). Flame lengths are c. 8 m in c. 200 cm tall vegetation, wind speed = 3 m s−1, RH = 16%.

Figure 7.

Erica stand canopy height vs. time-since-fire in the re-sampled transects. Stands denoted by × had burnt between the first and second inventory (one-year period).

Discussion

Fuel and Fire Behaviour vs. Time-since-fire

Substantial changes occurred in both quantity and structure of fuel with time since fire, and we identify three periods with distinct effects on fire behaviour: first, a period of around 5 years until canopy closure, where fire is blocked by lack of sufficiently continuous fuels; this is followed by a period where fuel accumulation follows shrub height growth and gives the potential for increasingly high fire intensity; finally, a period where canopy fuels become separated from surface fuel and fire potential decreases.

As little as 4 years after fire, the fuel load was 10 Mg ha−1, including a component of very fine dead fuel (< 1 mm), but nevertheless this did not support fire spread. Such fuel loads are well above the flammability threshold in Scottish ericaceous heath (Davies et al. 2009) which has a similar shoot architecture. We suggest that this difference is due to the discontinuous arrangement of the fuel in our study area, as Erica re-sprouts from large spatially separated lignotubers. A parallel example is mallee shrubland in Australia, where discrete fuel arrays have been shown to limit fire spread (Bradstock & Gill 1993). Controlled laboratory experiments suggest that individual fuel elements need to be ignited by direct flame contact (Finney et al. 2010), which in a discontinuous fuel bed would be facilitated by increasing wind speed, slope or fuel-bed depth (Finney et al. 2010).

The youngest burnt stand we recorded was 5 years old and c. 60 cm tall at time of fire, which appears to define the flammability threshold in grazed vegetation. After 5–6 years, the average height of Erica is 60–80 cm, canopy cover approaches 60% and more fine dead material has accumulated, also including slightly thicker dead Erica stems due to self-thinning, which are trapped upright in the canopy. Further, the moss starts coming back in larger quantities. Such stands may burn on steep slopes or under optimal weather conditions.

From c. 6 years, the live fuel components (< 6 mm diameter) approach a steady state, but dead material continues to accumulate and the horizontal and vertical continuity is good, resulting in a dramatic increase in fire-line intensity and flame lengths, particularly from 10 years onwards. The BehavePlus model is not constructed to handle spatially discrete fuel beds and is therefore of little use for the young stands, but for older closed-canopy Erica stands, the model produced estimates of fire behaviour comparable to the fires observed in the field.

When the Erica grows above c. 2·5 m, another threshold is reached as the flammable canopy starts to separate from the surface fuels. Without vertical connectivity between surface and crown fuels, fire spread is less efficient (Albini 1985), and according to local people, old stands are difficult to ignite (Johansson et al. 2012). If not burnt, the Erica eventually develops into a stunted forest, with even greater spatial separation between surface and canopy fuels (cf. Fig. 1d). Thus, the Erica heathland is highly flammable as long as it is burnt on short rotation, essentially maintaining the Erica in a ‘fire trap’ (cf. Bond & Midgley 2001). The minimum fire-return interval of about 5–6 years is controlled by fuel-accumulation rates, but should a stand remain unburnt long enough to develop into forest (> 30 years), flammability again decreases.

The role of fire in maintaining a sharp treeline on tropical mountains has been much discussed (e.g. Miehe & Miehe 1994; Bader, Rietkerk & Bregt 2007), but without reference to fuel structure and fire behaviour. We suggest that this border is controlled by differences in flammability, although with climate as an important contributing factor according to the following: as Erica growth rates decrease with elevation due to cooler temperature (Wesche et al. 2008), the time needed for stands to escape the fire trap increases and the chances of developing into a fairly non-flammable forest decrease. Once this border between forest and heathland has been established, it becomes increasingly stable with time, reinforced by the fact that the poorly flammable forest is always down-slope of the highly flammable heathland vegetation.

Effects of Herbivores on Fuel Succession

Fuel development over time was highly dependent on cattle herbivory, which affected the fuel situation in two ways. The first effect is that they consume young Erica shoots, slowing down the increase in height and canopy cover, thereby increasing the time needed to reach the minimum fuel load and spatial continuity for fire propagation. The large variation in browsed Erica stand height at any given age probably depends mainly on differences in browsing pressure, but site factors may be important too: possibly local climate and soil fertility (Johansson et al. 2010). According to the exclosure data, browsing may delay E. trimera with on average c. 3 years to reach the flammability threshold, thus increasing the longevity of fire breaks on the landscape. This represents a positive feedback, whereby herbivore activity maintains the stands longer in a favourable state for grazing/browsing.

The second effect of cattle on the fuel complex is that they graze the grass/herb vegetation in-between shrubs short, increasing fuel-bed heterogeneity in young stands. The short-cropped vegetation between shrubs, rarely exceeding 2 cm in height, cannot carry fire which means that the shrubs constitute the only available fuel in the system. An increased quantity of grass/herb biomass upon relaxed grazing could potentially provide an important fuel component, connecting dispersed shrubs in young stages, creating a more homogenous fuel bed that would allow shorter fire-return intervals. However, the quantity of grass/herb vegetation in the first years after fire inside exclosures was relatively small. Also, it would have to cure enough during the dry season to support fire and we never observed complete curing. Extreme drought years, which historically have occurred once a decade, could possibly be important in this respect (Wesche, Miehe & Kaeppeli 2000).

The cattle prefer young Erica and spend most of the time in 1- to 3-year-old stands (Gustafsson 2009), which may be typical for megaherbivores (Fuhlendorf et al. 2009; Augustine & Derner 2014). Therefore, the effect of browsing and grazing is greatest the first years after fire. In this early period, there is also an interesting legacy of the previous stand in that the standing charred dead stems (old-generation) evidently act as a browsing protection, but this effect should depend on stem size, that is, it should be related to fire interval/stand age at time of fire. After canopy closure, the grass/herb component more or less disappears and as the shrubs grow out of browsing height, which may be around 1·5 m (Sanon, Kabore-Zoungrana & Ledin 2007), the cattle no longer affect Erica fuel accumulation.

The effect of native herbivores on the heathland fuels is unknown. The main wild megaherbivore is the Mountain Nyala Tragelaphus buxtoni, a large antelope considered to be mainly a browser, but which has been observed to forage on both grass, herbs and young Erica shoots in fresh burns (Brown 1969). Today Nyala are present in very small numbers, but accounts of big hunts in the early 1900s suggest much larger populations in the past (Johansson, unpublished data). There are also different species of grass rats (SilleroZubiri, Tattersall & Macdonald 1995), which could have influenced grass growth inside our exclosures, but we did not observe much evidence of this. Pastoralism is ancient in the area (Miehe & Miehe 1994), but past cattle stocks have fluctuated greatly due to disease outbreaks, such as the Rinderpest panzootic in the 1890s (Pankhurst 1966). However, we expect none of the native herbivores to have as strong influence on fuels as the cattle, at least not during the last several decades.

Management Options

Under the current fire and grazing regime, fuel succession determines the shortest possible fire-return interval at c. 5–6 years, whereas the average fire-return interval at present is about 10 years, with a range from five to 19 years (Johansson et al. 2012). Shrub ecosystems dominated by vegetatively regenerating species are typically resilient to a wide range of fire intervals (Keeley et al. 2012), and the current fire regime seems to give the Erica shrubs enough time to recuperate between fire events. Nevertheless, changes in the fire regime as well as the browsing pressure would have consequences. We believe that the fuel limitation on minimum fire-return intervals makes it impossible to eradicate the Erica by fire alone, but a constant high browsing pressure could potentially weaken the Erica lignotubers over time (Paula & Ojeda 2011), exacerbated by short fire-return intervals because it results in smaller old-generation stumps that give less efficient protection to the newly emerging Erica shoots. Nevertheless, the Erica shrubs manage to put on c. 11 cm in height each year despite a high browsing pressure, probably because the cattle do not browse below the dense carpet of hardened shoot tips resulting from the previous browsing episode (pers. obs.). A larger threat to the Erica may be if it is burnt during extreme drought, because smouldering in the humus layer can then damage the upper part of the lignotubers (Johansson et al. 2012). Although seedling establishment does occur (Johansson 2013), long-term effects of loss of old lignotubers are unknown.

Burning and grazing increases the heterogeneity of the heathlands, and young stands harbour a wide diversity of open-habitat species which are absent in the more species-poor old Erica scrub (Wesche, Miehe & Kaeppeli 2000). Likewise, threatened endemic animals like the Mountain Nyala and Simien Wolf benefit from the increased heterogeneity as young Erica stands provide them more food resources than old stands (Brown 1969; SilleroZubiri, Tattersall & Macdonald 1995).

One important implication of the fuel limitation is that it allows the pastoralists to control the sizes of burns. The repeated burning creates a patchy landscape with numerous fire breaks. A substantial lengthening of fire-return intervals would thus have dramatic repercussions, first by eroding the grazing value and biodiversity of the heathlands and second by creating massive fuel build-up over vast areas, increasing the risk of intense large-scale wildfires that would be dangerous to both people and livestock (Butler & Cohen 1998). Such fires in turn would cause a momentary loss of grazing over large tracts, as it takes several months for the pasture to recover after fire (Johansson et al. 2012). The current situation where fire is officially banned but still practised locally does not favour sustainable land management, and this discrepancy between policy and practice might be further aggravated by upcoming carbon-storage plans. We recommend that a joint management plan is developed by authorities together with local pastoralists in order to achieve a management system that preserves these unique heathlands as well as the livelihoods of the pastoralists. This plan should allow fire but within defined prescription windows with regard to fire-return intervals, fire weather and grazing regime.

Fuel, fire and herbivores will interact in any fire-prone system, but so far the focus has mostly been on the first two factors, possibly because populations of high-impact herbivores are often tightly controlled by people. Our results suggest that interactions between all three should be simultaneously considered in land management discussions.

Acknowledgements

We thank our field assistants Ayano Abraham and Shebru Marefu. Sören Holm helped us with statistical analysis. Kristoffer Hylander kindly identified the moss species. Lisbet Holm-Bach and two anonymous reviewers provided valuable comments on the manuscript. The study was funded by the Swedish international development cooperation agency, Sida.

Ancillary