Fire and grazing controlling a tropical tree line: Effects of long- term grazing exclusion in Bale Mountains, Ethiopia

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2020 The Authors. Journal of Vegetation Science published by John Wiley & Sons Ltd on behalf of International Association for Vegetation Science 1Department of Ecology, Environment and Plant Sciences, Stockholm University, Stockholm, Sweden 2Department of Forest Ecology and Management, Swedish University of Agricultural Sciences, Umeå, Sweden

often have distinct boundaries, which suggest that other factors are more critical. Candidates would be competitive interactions among dominant plant species (Friis, 1986), vulnerable life history steps such as seedling establishment (Wesche et al., 2008) or the effect of disturbances such as fire or grazing (Bader et al., 2007;Oliveras et al., 2018). From climate alone, the potential for fire disturbance would be expected to decrease with elevation owing to increased precipitation and decreased evaporation. Instead, tropical mountains often have a higher frequency of fire above the tree line than below owing to a higher flammability of the alpine heathlands or grasslands (Wesche et al., 2000;Spehn, 2006;Crausbay and Martin, 2016). Numerous examples exist in South America (Ellenberg, 1979;Rehm and Feeley, 2015), Asia (Smith, 1980) and Africa (Hedberg, 1951;Wesche et al., 2000). Often, the fires are set by people to improve grazing and their spatial extent is controlled by post-fire fuel limitation, which prevents landscape-covering fires (Johansson and Granström, 2014).
Herbivory is a disturbance factor with some similarities to fire, but which more selectively removes plant biomass (Bond and Keeley, 2005). The most nutrient-rich and palatable parts of the plants will be selected, usually the youngest foliage (Davis, 1967).
Similarly, palatable species will be preferred over less palatable ones, altering their competitive balance (Johansson et al., 2009).
Fire regimes affect herbivory, by for example keeping vegetation short enough for grazing or browsing animals (Jáuregui et al., 2009) and by increasing the proportion of early-successional and palatable species. Indirectly, herbivory can also affect the fire regime by altering the composition of the fuel bed and the vertical connectivity of fine fuels, reinforcing or reducing a contrasting flammability between shrubland and forest (Johansson and Granström, 2014;Blackhall et al., 2017;Tiribelli et al., 2018). Thus, fire regimes and regimes of herbivory are interlinked, not least through culture, as people in many parts of the world supply both the bulk of the herbivores and the ignition of most fires.
On the high mountains of the East-African highlands, there is typically a sharp tree line at around 3,500 m with shrubby Ericaceous heathlands above a forest belt dominated by tree-shaped Erica species (Hedberg, 1978;Miehe and Miehe, 1994). Many Erica species, such as Erica arborea, are highly flammable because of oil-rich ericoid leaves, multi-stemmed shrub canopies and dead twig retention (Dehane et al., 2017). The species present here form lignotubers when exposed to frequent fires (cf. Paula and Ojeda, 2011). In the Bale Mountains, livestock grazing is the main land-use throughout the altitudinal gradient, but fire is more or less restricted to the zone above the tree line, where local pastoralists have burnt the heathland on short rotation for centuries (Gil-Romera et al., 2019) to keep it within browsing height (Johansson et al., 2012).
In order to analyse fire and grazing as controlling factors for the tree line, we did experimental and observational studies on the northern slopes of the Bale Mountains, Ethiopia. To elucidate the effects of grazing on vegetation dynamics and fire potential above and below the tree line, we constructed a set of grazing exclosures at six sites within a relatively restricted elevation range of ~150-250 m around the tree line. Fuel succession after fire in Erica heathland vegetation has been reported elsewhere (Johansson and Granström, 2014). Here, we analyse fire potential and vegetation colonization using experimental data collected mainly during the first five years after livestock exclusion. Potential for seedling colonization was assessed through seeding of the dominant woody species (Erica spp., Hypericum revolutum and Hagenia abyssinica) in the different vegetation zones. Further, we characterized soil properties and performed a soil bioassay using Hagenia.
An overall aim of the study was to identify whether the current fire regime and high grazing pressure would be sufficient to (a) control the surface fuel situation in the forest below the tree line and (b) to prevent tree colonization above the tree line. This would clarify the role of grazing in maintaining the contrasting flammability between the two vegetation zones, which in turn stabilizes the tree line at its current altitude.

| Study area
The study area is located on the northern slopes of Bale Mountains, Ethiopia, in a ~45 × 10 km area centred at N 6°50′, E 39°18′ ranging in elevation from ~2,950 m to 3,700 m ( Figure 1). The Bale Mountains has the largest area of subalpine and alpine vegetation in Africa and harbours many endemic and threatened species (Fetene et al., 2006). For a description of the geology and climate see the Supporting Information and Figure 1.
The vegetation shows a distinct altitudinal zonation (Friis, 1986;Miehe and Miehe, 1994 (Figures 1c and 7a), dominated by mixed stands of multistemmed Erica arborea and E. trimera shrubs (hereafter collectively referred to as Erica) in different phases of postfire regeneration, rarely exceeding 2-3 m in height before burning (Johansson et al., 2012). The heathlands are burnt on short rotation (average fire interval ~10 years, with stand sizes ranging from 0.5 to ~10 ha).
After each fire, the Erica shrubs quickly regenerate from large (~0.3-1.5 m diameter) buried lignotubers (spaced ~0.5-1.5 m apart) and young stands are dominated by a species-rich, grazed grass/herb field layer, hereafter referred to as the 'grass/herb lawn', in-between resprouting short Erica shrubs, creating a small-scale heterogeneity within stands (Figures 2 and 7d). Young stands are non-flammable the first four years because of the lack of fine fuels (Johansson and Granström, 2014). Throughout all zones, there is intense grazing by livestock, mainly cattle, with a few horses, sheep and goats (Johansson et al., 2012).

| Exclosure establishment
To quantify the effects of grazing, permanent livestock exclosures were erected in the three elevational zones: heathland, Erica forest and Hagenia forest. A summary of the set-up is found in Table 1. Sizes of exclosures differed (Table 1), but the construction was similar: a ~160 cm tall fence with six strands of barbed wire, 20 cm apart at the base, and 30 cm between the uppermost strands. Some sheep and goats were present close to habitations and here the exclosures were reinforced by vertical wood splints interwoven with the wire up to ~70 cm. Fences also excluded the Mountain Nyala (Tragelaphus buxtoni), but not rodents, monkeys or the high-jumping Bohor reedbuck (Redunca redunca).
All exclosures were guarded and maintained by local staff, and no signs of grazing or browsing were observed inside exclosures. The basic experimental design of all exclosures is a split-plot design: we randomly selected the placement of the fence, which was surrounded on all sides by a larger area of similar vegetation used as the grazed treatment.
In the heathland zone paired grazing exclosures (10 × 10 m) were established at three sites, with one exclosure in recently burnt Erica F I G U R E 1 (a) Location of the study area in Bale Mountains, in the Southern highlands of Ethiopia. (b) Climate diagram for the northern aspect tree line at 3,400 m.a.sl., average monthly rainfall, average daily T max (red line) and RH min (blue line) for 4 years (2006)(2007)(2008)(2009) Experimental burning was not allowed, so we made use of fires set by locals just before exclosure set-up. Therefore no pre-fire measurements had been made, but judging from remnant fire-killed stem heights and annual ring counts, all burnt stands had been of the same pre-fire heights (140-180 cm) and age (8-11 years) as the adjacent mature stand. Slope, soil and water conditions were also judged to be similar within each pair of recently burnt and mature stands.  (Table 1). Only the 2005 fire was directly observed by us, on February 23rd. All exclosures were installed shortly after fire, before new Erica shoots had emerged from the lignotubers. All sites had slightly sloping terrain (17%-34%) and contained both Erica species.
Sites were chosen to have moderate slopes and a closed E. trimera canopy, but a few Hagenia and Hypericum trees were present nearby. The trees were ~11 m tall and had two to six stems per tree.
Counting of annual ring counts on six trees (diameter 30-35 cm) revealed an age of ~ 90 years. The basal area was 30 and 38 m 2 ha -1 and canopy cover ~85%-95%. The field layer was an herb carpet cropped to 1.5-2.5 cm height by livestock (mainly cattle, horses and sheep). Control plots were surrounding the fences as described above.
At three Hagenia forest sites, ~14 and ~3 km apart, 30 × 30 m grazing exclosures were built in January-February 2006. Forest canopy and field-layer vegetation was similar inside and outside exclosures at the start of the study. The sites had a closed canopy dominated by 18-20 m tall Hagenia trees (diameter at breast height, dbh, 60-100 cm), with a few Hypericum trees. Basal area ranged 27-75 m 2 /ha, and canopy cover was 80%-90%. The field layer was a herb/grass carpet grazed to a height of 1.5-2.5 cm. Control plots were placed as above. At two Hagenia sites, we built an additional 5 × 5 m exclosure in an adjacent forest gap, <300 m away, with similar altitude, slope and soil, but with full sunlight, with surrounding controls. The forest gaps were 100-200 m wide and appeared to have been tree-less for a long time (no tree-stumps or other signs of recent clearing). They were covered by a 1.5-2.5 cm tall intensely grazed grass/herb lawn, with 50%-65% herbs.

| Vegetation sampling
In the heathland, vegetation cover was sampled using line transects inside/outside exclosures in burnt/mature Erica stands, giving four treatments ( Figure 2). Area cover of four different categories was Note: Experiment: 1) Vegetation development 2) Surface fuels (heathland fuels presented elsewhere) 3) Sowing/planting. a All fences were erected at the beginning of the year, in the dry season (January-February).
b Heathland fences were built in pairs, one in a new burn and one in adjacent mature Erica stand

| Heathland succession
To analyse effects of grazing on herbaceous post-fire colonization and seedling establishment, we established permanent field-layer vegetation plots (30 × 30 cm) inside and outside the heathland exclosures described above ( Figure 2). These were arranged in square blocks (2 × 2 plots) in-between the Erica shrubs or lignotubers, so that two plots were positioned directly adjacent to the lignotuber and two 30 cm out in the grass lawn. Plots were evenly distributed between the Erica shrubs, avoiding stones and tracks. To replicate the grazed controls (cf. Davies and Gray, 2015), the plots outside exclosures were distributed on three sides of the exclosures (Figure 2), and at a minimum distance of 3 m to the fence to avoid edge effects. In each block, one randomly chosen plot close to the lignotuber, and one further out, was scarified by mechanically removing the litter and upper 2 cm of the organic soil layer, including rooted herbaceous vegetation. Initial data analysis showed that position had little effect, and hence plot data from close/far from lignotuber were pooled. This gives a total of eight different treatments: 2 fire histories (recently burnt vs. mature Erica) × 2 grazing regimes (inside vs. outside exclosure) × 2 soil treatments (scarified vs. control), (n = 10 blocks at the 2005 site, n = 7 at the 2006 sites). All plots were marked by plastic sticks at the corners.
In these 30 × 30 cm vegetation plots, area cover of all plant species and non-vegetated soil/humus was visually estimated annually (until burnt plots were 3 and 2 years old, and mature plots were 14 and 10 years old, respectively). Erica seedlings were recorded as part of the field layer, but resprouting Erica branches were not. A total of 50 seeds of each species were sown on half of the plots of 1 × 1 m (but 0.5 × 0.5 in the gap exclosures, also sown with 50 seeds) with three different treatments, replicated five times: (a) control (no manipulation), (b) mechanical scarification (removal of herb/ grass vegetation) and (c) fire (leaf litter, ~400 g/m 2 was added to the plot and then burnt, see Hagenia burning experiments in Supporting Information). The fire treatment was excluded from gap exclosures.
No seedling germination was observed in the seeded plots in the Hagenia forests (presumably owing to shading from the lush herb carpet that covered the scarified plots in less than 6 months).
Additional planting was therefore done in June 2007 (rainy season) of 6 months old Hagenia seedlings, produced from our seed batches by a local nursery. At each Hagenia forest site, ~4 cm tall seedlings were planted along the fence ~2.5 m apart (n = 45 seedlings per treatment per site). At the two gap exclosures, nine seedlings were planted inside and outside. The herb carpet was then ~5 cm tall inside exclosures and ~2 cm outside exclosures. Seedling survival and height was monitored annually for 5 years and sapling height measured after 6 and 10 years.

| Soil sampling and Hagenia bioassay
To characterize soil properties (

| Heathland post-fire vegetation development
In the burnt heathland transects, there was rapid regrowth of Erica shoots from the lignotubers. Erica shrub cover outside exclosures was 21%-45% 1 year after fire ( Figure 3). Within 5 years, it was 40-70%, nearly on par with adjacent mature stands. Erica shrub cover increased significantly faster inside exclosures (p < 0.001, Appendix S1). After 5 years, it was 13-20%-units higher inside exclosures than outside (Figure 3). Cattle browsed the top ~10 cm of the Erica shoots, at a bite diameter of ~1 mm. The resulting dense carpet of woody stem-pegs appeared to restrict consecutive cattle browsing further down into the shrubs, and thus offered some protection (Figure 7e). The proportional cover of the herb/grass lawn increased rapidly and culminated 1-3 years postfire ( Figure 3). Thereafter, herb/grass cover declined, faster inside exclosures. Non-vegetated soil/humus decreased rapidly in burnt plots, and was virtually nil after 5 years, both inside and outside exclosures ( Figure 3).
In the 30 × 30 cm field-layer vegetation plots between the Erica shrubs, total herb and grass cover was on average 62% (SD = 18) 2 years after fire ( Figure 4). Few seedlings were observed (mainly of

| Forest field-layer vegetation and surface fuels
In the Hagenia forest, herb/grass cover ranged from 80% to 90% outside exclosures. Inside exclosures the herb/grass cover increased to

JOHANSSON ANd GRANSTRÖM
Grasses were rare in the Hagenia forests; only at one site did grass cover reach more than 20% of the field-layer vegetation inside the exclosure 2 years after fencing. After 10 years grasses covered more than 60% in one exclosure, while the other surviving exclosure still had a herb-dominated field layer (the third Hagenia forest fence was broken after 6 years).
In the E. trimera forest exclosures, the herbaceous layer had a similar species composition and attained ~ 10 cm height with a biomass dry weight (DW) of 141.5 g/m 2 (n = 8, SD = 31.8) 1 year after fencing. Here, almost all surface fuels consisted of a mesophytic herb carpet and there was no large airy leaf litter because Erica needles are small and form a relatively compact litter layer. Thin Erica twigs fallen from the forest canopy comprised ~18 g/m 2 DW fine surface fuels (n = 8, SD = 8.7). Outside the fence, the herb sward was ~1.5 cm tall with a DW of 64.4 g/m 2 (n = 8, SD = 12.3).

| Tree seedling establishment in heathland and forest
In the heathland sowing experiments, seedlings of all three tree species emerged (E. arborea, Hagenia, and H. revolutum) but in highly variable numbers ( Figure 6) In the gap exclosures, however, five and four (out of nine planted per exclosure) Hagenia seedlings survived. Some seedlings were evidently harmed by mole-rat digging the first years. Ten years after planting the largest surviving Hagenia trees were 5 m and 10 m tall

| Soil status
Soil pH was lower in the mature heathland than in Hagenia forest.
There was an overall negative relation between SOM and pH and SOM and bulk density, and fire only marginally reduced SOM and increased bulk density and pH (Figure 8a,b). In the bioassay, Hagenia seedling growth was lower in the heathland soils ( Figure 8c). Average seedling heights were ~50% higher per 1 unit increase in soil pH.

| D ISCUSS I ON
In the present study, manipulation of two putatively important fac-

| Heathland vegetation dynamics under fire and grazing
In the heathlands, fire was followed by fast recolonization from surviving vegetative structures in the soil, notably from the Erica lignotubers, but also from extant herb and grass rhizomes and roots.
Two characteristics of the fire regime might contribute to this: the relatively short fire intervals, allowing for a rich herb and grass component also surviving under maturing Erica shrub, and the typically shallow depth of burn (Johansson et al., 2012), allowing survival of regeneration buds in the humus layer. Similar patterns have been reported earlier for shrublands under regimes of short-interval fire (Céspedes et al., 2014;Parra and Moreno, 2017 (Vial et al., 2011), were attracted to taller grass vegetation inside exclosures. However, we did not observe any obvious signs of rodent grazing.
Livestock impacts on shrubland dynamics are highly variable and depend both on vegetation composition, livestock species and the duration of the land use. In Australian heathlands, introduced cattle had little effect on shrub fuels because the cattle avoided mature stands (Williams et al., 2006). In Patagonia, introduced cattle reduced Nothofagus shrubland flammability by reducing the amounts of fine dead fuels, but at the same time they prevented succession into less-flammable forest (Blackhall et al., 2017).
Incidentally, the same plant traits that allow for survival after fire, such as effective vegetative regrowth, common in our dominant species, would also promote high grazing tolerance (Canadell and Lopez-Soria, 1998;Bond and Keeley, 2005). Although we mainly observed vegetative regrowth, there was some seedling establishment of herbs, grasses and E. arborea, evidently from the large and species-rich seed bank (unpublished data).
In the present study, there were a few naturally germinated seedlings of E. arborea in the small plots, but they were efficiently removed by the cattle. Even inside the exclosures, Erica seedlings did not exceed a height of 5 cm after 5 years, probably owing to intense competition from the old Erica individuals, which had a head start resprouting from lignotubers. In facultative resprouters, there can be a trade-off between resprouting capacity, increasing competitive success under a frequent-fire regime, and a large seed production, increasing success after infrequent, but more severe fires (Paula and Pausas, 2008;Maia et al., 2016). In the Bale Mountain heathlands, Erica seedling recruitment is evidently minor owing to heavy grazing pressure and the fact that

| Fire and the formation of the tree line
There is a stark contrast in flammability between shrub-shaped Erica in the heathlands, and tree-shaped E. trimera in the zone below.
Excluding livestock for 5 years in the Erica forest did not increase surface fuel flammability. Even 10 years after fence construction the field layer was still dominated by non-flammable mesophytic herbs.
We suggest that the mechanism behind the formation of a distinct forest-heathland border is the reduced Erica growth rates with increasing altitude, owing to the harsher climate. This increases the time needed for the Erica shrubs to grow into tall trees and thereby escape 'the fire trap' (cf. Bond and Midgley, 2001). Once established, the tree line has been maintained at its current position because of the contrasting flammability of the heathland and forest zones.
It is probable that the tree line has been stable for a considerable time, perhaps since anthropogenic fire commenced, likely more than 2000 years ago (Umer et al., 2007;Gil-Romera et al., 2019). as Hagenia (Fetene and Feleke, 2001). If grazing would cease, germinants would have to start from under a dense herb field layer where there is even less light.

| Hagenia regeneration and potential for colonizing the heathland
In the heathland, where Hagenia seedlings indeed did emerge after seeding (mainly in scarified and burnt plots), the poor nutrient status, or the low soil pH, is instead the likely reason for their poor performance in the fenced plots as well. The nursery bioassay showed poor Hagenia growth in the acidic heathland soils, even at a more favourable climate at a lower elevation. In the heathland, Erica and mosses form a recalcitrant litter, producing a thick and acidic humus layer. This is in accordance with the scheme suggested by Read and Perez-Moreno (2003) of long-term soil-plant interactions in areas dominated by ericaceous plants. Most heathland fires in Bale occur when the humus is too moist to smoulder, and thus fire has little effect on the humus layer below the surface litter (Johansson et al., 2012). Of course, the harsh climate in the heathland might also contribute to poor Hagenia growth, but we have observed individual vigorous Hagenia trees well above the tree line, but always in fire-protected habitats with atypical soils: road banks or rocky outcrops (Figure 7b,c). The lack of basal resprouting in Hagenia (personal observation) would in any case eliminate the species under a regime of short-interval fire because natural fire refugia are rare in the heathlands.
It has been suggested that Hagenia would efficiently regenerate in the forest zone only after wildfire (Lange et al., 1997) because of its shade-intolerance (Fetene and Feleke, 2001), high palatability and efficiently wind-dispersed seeds. After wildfire, there would be more sunlight and perhaps also grazing protection from fallen fire-killed tree stems (cf. de Chantal and Granström, 2007)

F I G U R E 6
Seedling establishment in the heathland sowing experiment, 1 year after sowing. First-year establishment expressed as the proportion observed seedlings out of germinable sown seeds of Hagenia abyssinica, Hypericum revolutum and Erica arborea in the 30 × 30cm plots. Germination was higher on bare soil (burnt and scarified), survival was higher in mature stands (Error bars = 1 SE, n = 3 sites)  Mokria et al., 2017).
Our data show that relaxed grazing pressure is an absolute requirement for the regeneration of Hagenia, but this would also have to be accompanied by canopy opening. The few surviving Hagenia saplings in our gap exclosures (up to 10 m tall in 10 years, Figure 7f) testify to excellent growth potential in sunny sites, if grazing is excluded and the seedlings can escape competition with the field layer.

| Effects of livestock exclusion on stability of the tree line
Our results suggest that the distinct tree line, once established, is maintained by several factors which in turn stabilize a clear boundary with respect to flammability. The heathland is highly flammable, in contrast to the forest below because of differences in fuel structure and vertical fuel connectivity (Johansson and Granström, 2014 As for the highly flammable heathland, removing both livestock and fire would only slowly alter the fuel situation, or the chances for Hagenia colonization. It takes more than 20 years before the canopy fuels start separating from the surface fuels (Johansson and Granström, 2014) and substantially longer before flammability decreases. The other Hagenia barriers, the soil status and competition from resprouting Erica, are more long-term legacies that will remain, even if fire could be excluded (Read and Perez-Moreno, 2003). Similarly, excluding livestock from the forest will not alter flammability much. The poor surface fuels in both Erica and Hagenia forests will probably remain, as long as the forest canopy is closed, restricting the fuel bed to mainly leaf litter. In many forest, habitats reduced grazing of the field-layer results in increased surface fuel mass (Belsky and Blumenthal, 1997; Raffaele et al., 2011;Blackhall et al., 2015), but in our case the denser herb-dominated vegetation that followed on from livestock exclusion even resulted in decreased flammability, given the short dry period and low chances for field vegetation to cure.

| CON CLUS I ON S AND MANAG EMENT IMPLIC ATIONS
Repeated heathland burning over centuries has created a mosaic of Erica stands of different height and cover, resulting in rich biodiversity and extended habitat for Afro-alpine species (Johansson et al., 2018). There is palynological evidence of heathland dominance and anthropogenic fire during the last 2,000 years (Umer et al., 2007;Gil-Romera et al., 2019). Owing to a long history of burning and grazing, the Bale Mountains heathlands have remained in a fire trap. The tree line has not advanced upslope, as would be expected from the long-term warning trend in Africa (cf. Jacob et al., 2015). Within the framework of the new REDD + project in Bale Mountains, there are currently efforts to stop the burning and to impose stronger grazing restrictions (Watson et al., 2013). However, attempts to reduce livestock grazing are not likely to substantially change the present situation either with regard to fire potentials, or with regard to forest regeneration. The old forests will remain highly non-flammable and non-regenerating, and the tree line will remain in place. Heathlands will still be dominated by resprouting Erica and high flammability is likely to persist over the foreseeable future. Even without fire and grazing, the acidic soils, a legacy of centuries of Erica dominance can further prevent Hagenia advancing uphill into the heathlands.

ACK N OWLED G EM ENTS
Many thanks to field assistants Shebru Marefu and Ayano Abraham, the community forestry project and the local pastoralist community for enabling us to work in this remote area. We thank the anonymous reviewers for valuable suggestions.

AUTH O R CO NTR I B UTI O N S
Both authors were responsible for the conception of ideas and the study design. M.J. did most data collection, analysed the data, and led the writing of the manuscript. Both authors contributed critically to the drafts and gave final approval for publication.

DATA AVA I L A B I L I T Y
All data are submitted for archiving in Dryad with https://doi.