Savanna woody plants responses to mammalian herbivory and implications for management of livestock–wildlife landscape

1. The need to address wildlife conservation outside of protected areas has become more urgent than ever before to meet environmental and socio-economic goals. However, there is limited knowledge about how woody plants respond to herbivory within landscapes shared by wildlife and domestic herbivores in African savanna, thus management decisions might be based on inaccurate information and ulti-mately be ineffective. 2. We compared woody vegetation dynamics between two adjacent ranches with different management objectives and subjected to varying levels of herbivory by both wildlife and domesticated mammals using 421 square plots of 400 m 2 nested on three transects, each 3 km long and purposively selected to minimize bio-physical differences. 3. Both species and structural diversity were significantly higher ( p < 0.05) in the site with lower levels of herbivory. the site with higher levels of herbivory recorded enhanced biomass production a selection of palatable forage species, trees grow taller the is to promote co-existence of both wildlife and livestock, then the strategy would be to promote structural diversity by varying livestock stocking rate. Given the move-ment of wildlife between properties, livestock stocking rates should be considered within a wider landscape than just individual private lands.

is to promote co-existence of both wildlife and livestock, then the strategy would be to promote structural diversity by varying livestock stocking rate. Given the movement of wildlife between properties, livestock stocking rates should be considered within a wider landscape than just individual private lands.

K E Y W O R D S
biomass, East Africa, mammalian herbivory, semi-arid, stocking rate, structural diversity, wildlifehuman landscapes

INTRODUCTION
Savanna systems cover more than half the area of the African continent and support a large fraction of its human population, majority of its rangeland and livestock biomass and are home to the greatest density of wild herbivores and carnivores of any ecosystem on Earth (Sankaran and Anderson, 2009). In addition to their obvious socioeconomic importance, several savanna grazing systems suffer from woody/bush encroachment in both Southern Africa (Moleele et al., 2002;Roques et al., 2001;Trollope et al., 1998) and Eastern Africa (Angassa, 2012;Dalle et al., 2006). Bush encroachment has been associated with decline in grass resources for livestock (Macharia and Ekaya, 2005;Moleele et al., 2002), decreases in soil carbon (Hudak et al., 2003) and declines in biodiversity (Angassa, 2012;Gordijn et al., 2012). Several causal factors have been brought forward to explain mechanisms behind woody plant encroachment and key among them include changes in: the intensity of grazing, fire, browsing and rainfall (Angassa, 2012;Archer et al., 2017;Dalle et al., 2006;Roques et al., 2001). The role of browsers in driving woody plant dynamics however remains unclear. A study by Mills et al. (2005) in South Africa noted that intensive goat production transformed semi-arid thicket with dense shrubby vegetation into open landscape dominated by ephemeral species. Conversely, Roques et al. (2001) observed that browsing pressure had significant but minor impact on the woody plant dynamics only at the early stages of encroachment. Earlier studies ascribed bush/woody encroachment to selective ungulate herbivory that led to a proliferation of unpalatable and/or chemically defended species that outcompete defoliated palatable species; however, other studies have also shown that grazing/browsing-tolerant palatable species often increase with increase in herbivory (Augustine and Mcnaughton, 1998;Hulme, 1996;Skarpe, 1990).
In Kenya, a number of landscapes under varied land uses have equally suffered from woody plant encroachment. In northern rangelands (especially Laikipia County), bush encroachment has been observed largely on private commercial ranches (Augustine, 2003a;Okello et al., 2001;Wahungu et al., 2013) and some communal pastoral lands in the southern rangelands (Kidake et al., 2015;Macharia and Ekaya, 2005). Both private and communal ranches have lost critical ecosystem services in northern rangelands such as pasture availability, ability of the ecosystem to resist invasion by both native and non-native species, loss of wildlife habitat among other ecosystem services. Communal ranches in Laikipia County are particularly vulnerable to pasture scarcity compared with private ranches within similar bioclimatic conditions due to land degradation, invasive alien species and woody plants encroachment (Kimiti et al., 2020;Strum et al., 2015). The most common encroaching species in Kenya rangelands include: Tarchonanthus camphroretus, Dodonaea viscosa, Acacia reficiens, Ipomoea kituensis, Sansevieria volkensii, Euclea divinorum among others (see Augustine, 2003a;Kidake et al., 2015;Kimiti et al., 2020;Wahungu et al., 2013). Increase in camel and goat numbers in the region in the recent past is thought to be a strategy to take advantage of increasing browse and/or cope with climate variability (Huho et al., 2011;Vehrs, 2016 giadis, et al., 2007). Deeper understanding of browse-browsers interactions is therefore needed so as to inform management strategies.
This study aims at determining the influence of different level of mammalian herbivory on woody species composition, structural diversity and browse biomass production to inform management of livestockwildlife-dominated landscapes.

Study site
The study was conducted in Mpala private commercial ranch (hereafter refer to PR) and adjacent Il Motiok communal pastoral group ranch (hereafter referred to as GR) in Laikipia County, Kenya centred at 37 • 53′E and 0 • 17′N in a semi-arid savanna ecosystem ( Figure 1).
The area receives mean annual rainfall of 450-600 mm in a weakly trimodal pattern with rain expected in April-May, August and October with January-March being a dry season (Augustine and McNaughton, 2006 (Cronk, 2002;Fox, 2018 (Kibet et al., 2016). Despite this change in the land tenure system, the management of communal ranches remained unchanged until the early 21st century. Since then, there has been land use changes such as increase in the number of camels, establishment of community wildlife conservancy, mining of sands along dry river beds and most recently subsistence cultivation along Ewaso Ngiro River (Kibet et al., 2016).
The area is dominated by fine-leaved species of Acacia genus; Acacia etbaica Schweinf., Acacia brevispica (Harms), A. tortilis (Forssk.) Hayne, Acacia drepanolobium Sjostedt and A. mellifera (Vahl) Benth (Young et al, 1995 tained low livestock stocking rate of 10-12 TLU/km 2 mainly cattle, camels and sheep over the past 3 decades. The GR stocking rates on the other hand fluctuate depending on prevailing weather conditions with high stocking rates during rainy seasons when pastures and water are plenty and low shortly after major droughts. The GR is estimated to be stocking at higher rate than PR (see Georgiadis et al., 2007;Kayezwiebel and King, 2014;. In the GR, the grazers kept included cattle, sheep and donkeys and browsers include goats and camels. Average estimates of combined grazing and browsing wildlife and domestic animals in the region  indicate that GR had densities of 43 TLU/km 2 compared with 28 TLU/km 2 for PR. Domestic and wildlife browsers, considered the most effective in modifying woody plant structure and perhaps its composition, were estimated at 28 and 17 TLU/km 2 for GR and PR, respectively (Augustine, 2003b;Kayezwiebel and King, 2014;Ngene et al., 2013).
To corroborate on the use of fire as a management tool in the study sites in the last 40 years since the group ranches were established, 10 key informant interviews targeting senior persons who have lived in the region for more than 35 years were conducted. Physical checks for burn stumps as well as burnt scars from older trees were also done.

2.2.2
Sampling design for structural data kilns and trees harvesting were avoided and alternative nearby site was selected. Twenty subplots were discarded in PR due to wildlife threats during fieldwork. In each subplot, percent vegetation cover, bare ground, percent slope, elevation and soil texture were recorded.
Soil texture was determined in the field based on feel flow chart protocol (see Vagen et al., 2010).
In each subplot, all woody species with stem diameter of 1.0 cm at approximately 50 cm above ground (hereafter referred to as diameter at knee height -DKH) were enumerated. Bedside DKH, height, canopy depth and canopy diameter (CRWN) were measured and recorded.
Non-woody invader species were also recorded and their abundance cover estimated. Botanical nomenclature followed Flora of Tropical East Africa (1954). Duplicate copies of specimen for each species were collected for re-distribution between East African Herbarium and Mpala Research Centre.

Structural diversity
Structural diversity and biomass production estimates were based on four dominant palatable species, A. brevispica, Acacia tortilis, Acacia mellifera and A. etbaica based on Lusigi et al. (1984) rating and from interviewing knowledgeable local herders. A. mellifera and A. etbaica overlapped in PR and GR and therefore provided basis for comparing browsing effects at species level.
Canopy volumes, canopy area and canopy densities per subplot were then calculated based on structural data collected using the formulas provided below.
where D 1 and D 2 are the two perpendicular diameter measurements when projected on the ground. This formula does not assume symmetry of the canopy.
Based on ellipsoid volume formula, where H d is the length of canopy depth, while D 1 and D 2 are the two canopy diameter readings (Thorne et al., 2002).
This is summation of individual trees canopy area (Equation 1) divided by subplot area of 400 m 2 then multiply by 100 to make it into a percentage (Manila, 2007).

Coefficient of variation (CV) =
where is standard deviation and is the mean.
Vegetation piospheric effects were tested in GR with apparent grazing/browsing gradient based on two focal points; settlement area (homesteads) and a temporary watering point at the other extreme end. Often livestock spend more time in early mornings, late afternoon and night near homesteads similarly at watering points during the day.
The two focal points were 3 km apart. We hypothesized that higher browsing and grazing occur near homesteads and close to watering points. To test this hypothesis, we compared tree densities, tree canopy area and percent bare ground using 20 × 20 m subplots along the transects. It was anticipated that lower tree and canopy densities and higher percent bare ground on subplots near homesteads and at the watering point compared with subplots in the middle of the transects.

Browse biomass
The browse biomass was estimated using double sampling method as described by Foroughbakhch et al. (2008). brevispica (r 2 = 0.24) and absence of A. tortilis in sampled plots in PR, the two species were therefore excluded in biomass estimates discussions.

Soil measurement
To isolate effects of soil properties, 45 soil samples were collected per site picked in a stratified random manner from the subplots described above. Plots were stratified by soil formation (black cotton, transition and sandy) and randomly selecting subplots within each formation. In each subplot, five subsamples were augured 0-30 cm deep from four corners and at the centre and lumped into a composite sample. The composite samples were sun dried and later transported in labelled ziplock bags to National Agricultural Research Laboratories, Nairobi for further processing and analysis. Standard methods were followed in pH and macro and micronutrients analysis as indicated.
Available nutrient elements (P, K, Na, Ca, Mg and Mn). The Mehlich Double Acid Method was used (Mehlich, 1984). The oven-dried soil samples were extracted in a 1:5 ratio (w/v) with a mixture of 0.1 N HCl and 0.025 N H 2 SO 4 . The elements Na, Ca and K were determined using a flame photometer and P, Mg and Mn using a spectrophotometer.
Total organic carbon. Calorimetric method was used (Murphy and Riley, 1962): All organic C in the soil sample was oxidized by acidified dichromate at 150 • C for 30 min to ensure complete oxidation. Barium chloride was then added to the cool digests. After mixing thoroughly digests were left to stand overnight. The C concentration was then read on the spectrophotometer at 600 nm.

RESULTS
The key informants unanimously confirmed that the use of fire as a management tool has not happened in the region in over 30 years; however, accidental fires have occurred in both sites but over very small extents particularly in Mpala ranch. The sampled plots were however not affected based on both the interviews and field observation, with no burned stumps observed in any of the sampled plots. A total of 7190 woody plants were enumerated; 1619 in GR and 5571 in PR. Species diversity varied between sites (Shannon index 1.95 and 1.01 for PR and GR, respectively). Forty-five species belonging to 24 families and 33 genera were identified in the study sites (Table 1). Private ranch with fairly low intensity of browsing had higher structural diversity but low biomass for selected palatable species compared with group ranch.

Environmental effects
To isolate possible effects of nutrient, percent slope and elevation, we undertook CCA ordination for measured variables. Observation from the ordination at GR showed that soil fertility declined with increase in distance away from homestead given that all macro-nutrients were Higher species diversity was observed in areas higher elevation gradient and with lower soil nutrients. Areas dominated by black cotton soil were low in species diversity.

Community woody structure
Density of woody species between sites varied significantly (p < 0.05) with mean of 739 and 180 trees per ha for PR and GR, respectively.

Piospheric effects
Using distance as a proxy indicator of grazing and browsing intensity (see Wesuls et al., 2013), we analyze tree densities, percent bare ground and canopy area along the transects. Descriptive statistics indicates low tree densities near homesteads followed by progressive rise numbers with increasing distance until about 2.7 km where a downward trend was observed (Figure 5(a)). High percent bare ground near homesteads and near watering point and less in subplots at the mid-  Figure 5(b)). Tree canopy area on average was low for subplots closer to homesteads with the highest canopy area occurring around 0.6 km distance from homesteads thereafter a steady decline towards the watering point was observed (Figure 5(c)).

Structural diversity
GR had significantly higher structural mean values in all variables measured (Table 3); however, in terms of structural diversity, the site was less diverse. High structural diversity as indicated by high values of CV was observed in basal area, canopy volume, canopy area and DKH, while height, canopy diameter and canopy depth showed low structural diversity in PR and GR.  Table 4).

F I G U R E 5
Piospheric effect of increase intensity of grazing and browsing along transects starting near homesteads at and ending close to watering point at 3 km creating two focal points at extreme ends. Changes along the transect using means values for (a) tree densities, (b) percent bare ground and (c) tree canopy areas The U-test significant values are indicated in bold. Note A. tortilis and A. brevispica did not occur across the ranches.

F I G U R E 6 Combined vertical canopy volume for selected forage species between PR (Solid bands) and GR (hatched bands) in Laikipa County, Kenya
We compared canopy volume for PR and GR and noted similar trends as those in Figure 3. PR had more canopy volume in lower height classes, while GR had higher values for tree taller than 3 m ( Figure 6).

DISCUSSION
This study shows that mammalian herbivory, in association with abiotic factors, influence plant species diversity, structural diversity and biomass production as an ecosystem process. Heterogeneity in vegetation structure influences functioning of African savannas, such as for-age production as well as providing habitat for a wide variety of plants and animals.

Species composition
The intensely browsed site (GR) had fewer woody species compared with PR although they shared 26% of the same species. Often herbivore species selection is low under higher browsing intensity and therefore both palatable and less palatable species are browsed (see Augustine and Mcnaughton, 1998) and under such circumstances, browsing-tolerant species will thrive at the expense of less tolerant species. Low species diversity in GR therefore may suggest a case of filtering out of species intolerant to heavy browsing or declining competitiveness resulting in low reproduction ability as noted in studies by Díaz et al. (2001) and Fornara and du Toit (2007). Although the study sites had fairly similar biophysical properties, A. drepanolobium with more than 1600 individuals in PR was represented by only two individuals within GR. This could have been linked to limited successful recruitment of younger individuals into the system. A study by Goheen et al. (2007) observed that resource re-allocation to defence instead of reproduction occurs under heavy browsing and this decreases recruitment of seedlings. Low numbers of the species could be due to recruitment bottlenecks that could be both biotic and abiotic or both.
Notably, species with high abundance in the GR, except one, were of medium to high palatability based on Lusigi et al. (1984), implying that these species are either browsing tolerant and/or browsing resistant. This implies that intense herbivory not only facilitates an upsurge of unpalatable species as noted by Riginos and Hoffman (2003), but also promotes proliferation of browsing-tolerant palatable species.
Encroachment by palatable species into rangeland is therefore less obvious and less problematic for pastoralists given their forage value and therefore its negative ecological effects such as declining wildlife habitat will go undetected until serious impacts are realized. Interviews with local community at GR did not classify A. mellifera as an invasive/encroaching species because the word 'invasive' has a negative connotation, and the species was not perceived as negative.
High occurrence of encroaching species such as Sansevieria sp, I.
kituensis and emergence of A. reficiens in GR suggest potential increase in resource availability associated with low intensity of competition and/or existence of unoccupied ecological niches (see Alpert et al., 2000;Schellberg and Pontes, 2013). Similarly, we suggest that the decline in I. kituensis abundance with increasing distance from homesteads is a function of soil fertility and/or disturbance (trampling). The species has been shown to thrive in overgrazed sites (Kidake et al., 2015;Macharia and Ekaya, 2005).

Woody community structure
In the absence of fire within the study sites in recent decades, we suggest that soil, climate and herbivory are possible vegetation determinants within this savannah ecosystem. Differences in herbivory levels seemed a plausible regulator of vegetation dynamics within the study sites given similarities in climatic and edaphic factors. Low woody plant density of 180 trees/ha in GR compared with 706 trees/ha at PR suggest possible poor recruitment of seedlings over the years. Size class distribution analysis for A. mellifera and A. etbaica (see Figure 3) indicated fewer seedlings and mature individuals and more individuals in middle size classes. Possible explanation to such observation could include: low seed development associated with declining fitness of heavily defoliated plants (see Fornara and du Toit, 2008); increase reallocation of resources to defence instead of reproduction (see Goheen et al., 2007); or death of younger individuals attributed to heavy browsing and trampling by browsers (see Augustine and Mcnaughton, 2004;van Langevelde et al., 2003). The last argument is most plausible within the study site. Trees below 3 m are accessible to majority of animals (sheep from ground level up to 0.87 m, goats at up to 2 m and camels up to 3 m high); this means shorter individual trees would suffer more defoliation than taller ones (Rutagwenda et al., 1989). A. mellifera individuals differed significantly at size classes below 3 m but less above 3 m. Augustine and Mcnaughton (2004)  Higher intensity of herbivory in GR and/or low species diversity could be responsible for relative homogeneity in vegetation architecture at GR. On the other hand, higher heterogeneity observed in PR may be a factor of selective herbivory and/or presence of higher species diversity with diverse lifeforms -trees and shrubs. High livestock stocking rate reduces species selection during feeding and this may promote establishment of 'browsing lawn' (see Cromsigt and Kuijper, 2011;Fornara and du Toit, 2007

Edible biomass production
Biomass production per unit plant as well as unit area was significantly higher in intensely browse site (GR) and could be linked to both biotic and abiotic factors. This observation was particularly pronounced among A. mellifera individuals below 3 m high. This was demonstrated by a greater number of stems per ramet and wider canopies associated with breaking up of apical dominance and increase in growth of lateral buds. Importantly, biomass of individuals taller than 3 m did not differ significantly between sites. Except for camels, most domestic herbivores browse below 3 m (Rutagwenda et al., 1989) and therefore taller individuals within GR, even though subjected to higher stocking rate, escape intense browsing. Compensatory re-growth among browsing-tolerant species has been noted in other studies; A. tortilis by Oba and Post (1999); A. nilotica and A. karroo (Sebata, 2013); A.
nigressens (Fornara and du Toit, 2007) and Indigofera spinosa by (Oba et al., 2000). Browsing enhances biomass production of browsingtolerant species through increased photosynthesis, cell division and reduction in rate of leaves senescence (Sebata, 2013). This implies that palatable species are able to survive intense browsing by strategically investing resources for faster growth to escape the 'browse trap' -browsing may impose height-structured recruitment limitations on trees (see Staver and Bond, 2014). We suggest that the formation of an hourglass-shaped architecture is one such strategy where browsed saplings initially form a 'browsing lawn' at the height of about 1 m that spreads laterally thus creating a buffer/safe zone for shoots at the middle of a ramet to grow and achieve a height beyond 3 m (Figure 7). Once the trees are fully established the lower branches senescence out and the individuals escape browsing trap altogether in the absence of larger herbivores such as elephants and giraffes. Interestingly, there was no significant variation in biomass above 3 m for A. mellifera for the two sites. This suggests that compensatory photosynthesis necessary to replace loss biomass is no longer tenable at that height ( between biomass values and numbers of individuals and canopy volume per hectare was observed in both sites.

CONCLUSIONS
1. Herbivory modifies vegetation directly through alteration of composition and structure and also indirectly through modification of growing environment such as enhanced nutrients levels, trampling and compaction of soil near kraals and watering points.
2. Intense browsing is thought to alter species composition whereby browsing-sensitive species decline and browsing-tolerant species increase.
3. Browsing of browsing-tolerant species causes the establishment of a temporary browsing lawn where browsed individuals quickly replace loss biomass, increase biomass production and promotes re-browsing. Oddly, the species also enhances the plants defensive strategies thus causing pain to herbivores at the same time benefit from inflicted pain from defoliation. We believe that growth vigour associated with browsed individuals is part of a strategy to escape herbivory given that compensatory growth ceases once the individuals escape the 'browse trap' zone.
4. To increase forage for livestock within the landscape would demand that browse-tolerant species canopies be 'maintained' within browsing height of the main herbivore species kept. However, if the management objective is to promote co-existence of both wildlife and livestock then, the strategy would be to promote structural diversity by varying livestock herbivory intensity.

ACKNOWLEDGEMENTS
Financial support was provided by FOREST-GEO grant and STRI/Levinson award for which SK is sincerely grateful. We received valuable support from a team of Research assistants lead by Kimani

CONFLICT OF INTEREST
The authors declare no conflict of interest.

PEER REVIEW
The peer review history for this article is available at https://publons.