Effects of short-duration kraaling depend on initial conditions in a mesic grassland

Short-duration overnight kraaling has been suggested as a tool for restoring degraded rangelands. However, the response of different plant functional types and communities to such intense livestock impact may vary depending on local context. We thus examined the effects of short-duration overnight kraaling on soil and vegetation characteristics in a mesic montane grassland in South Africa using paired kraal and control sites, as part of a low intensity grazing approach. Kraaling increased soil P and S, as well as soil organic matter (except when initial values were over 12%). The effect of kraaling on vegetation was strongly dependent on initial condition. Basal cover of grasses and forbs increased by approximately 50 and 15%, respectively, if sites had very low initial basal cover, but decreased by up to 15% if initial values were over 50% and 10%, respectively. Kraaling always decreased herbaceous biomass, but especially when initial values were over 2 000 kg ha−1. In mesic grasslands, short-duration overnight kraaling is promising as a tool for rehabilitating degraded sites or fertilizing abandoned cropland, but should be avoided where the grass sward is intact. We recommend that the suitability of kraaling be evaluated per vegetation type and local context.

The use of short-duration overnight kraals (corrals or bomas) has received considerable attention in the last decade as a potential tool for restoring degraded natural rangeland vegetation (Huruba et al. 2018), sometimes in the context of Holistic Planned Grazing (Peel and Stalmans 2018). Short-duration overnight kraaling systems have been adapted from the traditional long-term cattle corralling, which was used as a way of protecting livestock against predation and theft, and can result in distinct grass patches dominated by grazing-tolerant grass species and high concentrations of nutrients (Blackmore et al. 1990;Porensky and Veblen 2015). Short-duration kraaling is promoted as stimulating grass recruitment and growth through a variety of mechanisms. In bare patches, trampling breaks the soil surface and is claimed to increase water infiltration. Hoof action is thought to incorporate dead grass material into the soil, increasing soil organic matter and making light available for grass growth and seedling emergence (Sibanda et al. 2016). Kraaled livestock can redistribute and concentrate nutrients and grass seeds by grazing in the surrounding vegetation and excreting in the kraals (van der Waal et al. 2011).
Experimental studies illustrate that the effects of kraaling differ between vegetation types and with the type and density of the herbivore communities utilising the vegetation (Porensky et al. 2013;Sibanda et al. 2016). In semi-arid savannas in Zimbabwe, grass nutrient content and diversity increased with short-duration kraaling of livestock, with a shift towards palatable, grazing-tolerant species (Sibanda et al. 2016;Huruba et al. 2018). These grasses mostly established from seed on ground that was bare immediately after kraaling (Sibanda et al. 2016). Peel and Stalmans (2018) found that kraaling cattle at high densities resulted in the tuft size of perennial grasses decreasing, but overall basal cover increasing over a five-year period in the dry savanna of western Zimbabwe. Huruba et al. (2018) found that basal cover decreased with kraaling, but this was likely an indirect effect caused by intense grazing by warthogs at the abandoned kraal sites. In a Kenyan savanna utilised by high densities of wildlife and cattle, single short duration kraaling events resulted in persistent nutrientenriched patches dominated by palatable, grazing tolerant grasses. These patches were maintained for up to three years after kraal abandonment by high densities of wildlife preferentially grazing there (Porensky and Veblen 2015). In contrast, at a more lightly grazed site, the stimulating effects of short-term kraaling on soil nutrients and grass productivity were transient (Sibanda et al. 2016).
The effect of kraaling on plant cover, production and composition also depends on the functional traits and

Effects of short-duration kraaling depend on initial conditions in a mesic grassland
life histories of the dominant plant species. In arid Karoo vegetation in South Africa, trampling during short-duration kraaling had no effect on grasses, but decreased the diversity and cover of long-lived shrubs (McManus et al. 2018). Besides shrubs, excessive trampling can damage growing tips of grasses, especially long-lived tussock grasses (Dunne et al. 2011). The intense disturbance caused by overnight kraaling is likely to favour plants with life histories tolerant of disturbance, such as grasses with stoloniferous or prostrate growth forms, tough wiry grasses, or short-lived species that produce large amounts of seeds. In contrast, mesic grasslands in South Africa are dominated by long-lived tuft grasses and forbs with relatively low turn-over rates (Cingolani et al. 2014;Chamane et al. 2017), and many have low rates of reproduction from seed and are poor colonisers (O'Connor and Pickett 1992;O'Connor 1994;Everson et al. 2009;Everson et al. 2021). Thus, it is likely that mesic montane grasslands have different dynamics to semi-arid savannas and will respond to intensive animal trampling in different ways.
Many mesic montane grassland areas in South Africa are vulnerable to loss of grass cover and soil disturbance, for example through ploughing or heavy continuous grazing, which has resulted in bare ground and severe soil erosion in many areas (Vetter 2007). Long-term kraaling in these vegetation types has led to localised reduction in perennial grass cover and diversity (Morris 2017;Shezi et al. 2021). Invasive woody species, such as wattle, pines and eucalypts have also led to a loss of grass cover and large bare patches remain after clearing of invasive vegetation. Anecdotal evidence (Nicky McLeod, Environmental and Rural Solutions, 2015, pers. comm.) suggested that short-duration kraaling can lead to increased grass cover on such bare patches and thus be a potential restoration tool. However, before short-duration kraaling is encouraged in mesic grasslands, it is important to determine experimentally whether it is effective in restoring grass cover on bare soil, and whether it poses a risk of reducing grass cover in areas that have intact grass swards. The aim of this study was to examine the effects of short-duration overnight kraaling on herbaceous vegetation and soils in a mesic montane grassland, in the context of low-intensity, two-camp grazing, namely not an intensive Savory approach (Savory 2013;Savory and Parsons 1980). We wished to determine whether the magnitude and direction of the effects was influenced by initial conditions (basal cover and bare ground, soil conditions, slope, the previous clearing of wattle, rainfall immediately before and during kraaling) and the intensity (livestock units and days) of the kraaling event.

Study area
The kraaling experiment was done between 2014 and 2016 in Motseng (30°17′45″ S, 28°22′08″ E) and Mvenyane (30°32′13″ S, 29°01′32″ E) grazing areas in the Matatiele Local Municipality in the Eastern Cape, South Africa. The long-term (2012 to 2018) mean annual rainfall and temperature is 710 mm and 15.9 °C, respectively, with most rainfall occurring during austral summer months (WorldClim v2.1 database (Hijmans et al. 2005)). The area lies in the southern reaches of the Drakensberg mountain range at altitudes ranging from ca. 1 300 to 1 600 m asl with moderate to steep slopes. The landform and soil type based on SoilGrids (Hengl et al. 2017) are predominantly steep lands with Regosols, where the northern study sites (2 to 9, 12) are high-gradient mountains with Leptic Regosols and the southern sites (11,13,14) are mediumgradient mountains with Eutric Regosols (Table 1). The steep northern sites had a gradient of more than 30%, a relief index greater than 300 m km −2 and a drainage density of 0 to 15 potential drainage density (PDD). The less steep southern sites had a gradient of 15 to 30%, a relief index of 150 to 300 m km −2 and a drainage density of 0 to 15 PDD (van Engelen and Dijkshoorn 2013). The vegetation of the study area comprises three grassland types (Table 1) characterised by forbs, bunch grasses including Themeda triandra, Tristachya leucothrix, Heteropogon contortus, Eragrostis curvula and Andropogon appendiculatus, dwarf shrubs, such as Chrysocoma ciliata and Pentzia cooperi, and larger shrubs and trees including Leucosidea sericea and Vachellia (formerly Acacia) karroo (Mucina and Rutherford 2006).
The study area is communal grazing land. Similar to other communally grazed montane grassland areas in the Eastern Cape (Vetter 2006;Vetter et al. 2007), the study area is largely characterised by high stocking rates (up to 1.5 LSU ha −1 where 0.25 LSU ha −1 is recommended), continuous grazing of cattle, sheep and goats, little coordinated grazing management, and widespread soil erosion and vegetation loss (i.e. degradation). In addition, invasion by exotic woody species, such as Acacia mearnsii, has resulted in localised loss of grass cover and changed soil characteristics when canopy cover is high, with little or no recovery of grasses after clearing, possibly due to an allelopathic effect (Souza-Alonso et al. 2013).

Experimental design and data collection
Twelve paired sites comprised kraaled and adjacent non-kraaled (control) plots (Table 1). Overnight kraaling took place once for 7-24 d at the twelve paired sites between 2014 and 2016 within two designated grazing areas, as part of a low intensity, two-camp grazing approach. Between sunrise and sunset animals were herded by shepherds to one of two designated unfenced grazing camps outside the villages and allowed to freely graze on native grasslands as part of Conservation International's 'Herding for Health Programme'. Shepherds stayed nearby the overnight kraals in temporary tented stock posts that moved between grazing camps over the austral growing season (November-April). Post-kraaling sampling of soil and vegetation parameters was done in at each site in the paired kraal and adjacent control plots in 2016. For kraaling, herds of between 123 and 250 cattle and 115 to 286 small stock were kept overnight in movable circular corrals (kraals) approximately 50 m 2 in size and constructed from poles, netting and electrified wire. Because the experiment was part of a working development project, kraaling times and locations were determined in collaboration with the livestock owners from local communities and according to the needs of the livestock and weather conditions. A severe drought started in 2015, which terminated the kraal experiment during 2016 at the request of the livestock owners. Of the twelve kraal sites, eight were previously cleared of wattle and four were on untransformed grassland. Unplanned fires took place in 2014 and 2015, and four sites (all previously cleared of wattle) were burned since kraaling.

Soil analyses
Five soil samples were collected at each kraal and control site using a soil auger (20-cm depth, 10-cm diameter). Soil samples were air-dried, 2-mm sieved and subsequently stored in sealed plastic bags prior to laboratory analysis. Soil samples for elemental analysis were ground with a mortar and pestle prior to X-Ray Fluorescence (XRF) analysis to determine total elemental concentrations in soil samples (SPECTRO XEPOS XRF Analyzer, Kleve, Germany). Soil organic matter (SOM) was measured as loss on ignition on 5-g subsamples of the 2-mm air-dried soil, and the weight losses at 105 °C, 550 °C and 1 000 °C were recorded to estimate soil residual humidity, organic matter, and carbonates and residual water, respectively (Heiri et al. 2001).
To measure the infiltration rate of water into soil, a mini disc infiltrometer (Meter Group Inc, USA) was used at each paired site. The suction rate was set to three for all plots, because the soil types were similar. The infiltrometer was filled with water and placed flat on the soil surface after removing litter and other organic matter. Infiltration was calculated based on infiltration models (Zhang 1997) using the average volume of water lost to the soil every 60 s (n = 5 per kraal and control site).
Soil bulk density, representative of the middle of a 20-cm soil core, was measured using a steel ring (height 8 cm, volume 79.48 cm 3 ) that was gently hammered into the soil at the 8 to 13-cm soil layer in each control and kraal pair, and the contents carefully removed to avoid disturbing the sample. Any plants or roots were cut off at the ring surface with scissors. The soil was emptied into plastic bags and sealed. Soil samples were weighed on the day of collection, large stones removed, volume adjusted and then oven-dried at 105 °C to constant weight and re-weighed. Bulk density (g cm −3 , fine fraction) was calculated as dry soil weight (g) per adjusted soil volume (cm 3 ) after subtracting the volume of the coarse fraction.

Vegetation analyses
The percentage basal cover of dominant forbs and grasses and of bare ground was estimated using a 'Levy bridge' or movable frame (Everson et al. 1990) with 10 drop pins 25 cm apart, dropped 10 times to sample 100 points per kraal or control plot. A plant species was recorded only if its rooted live base was touched by a dropped pin. Bare ground was recorded if the pin did not strike the rooted base of a plant. Grass species were identified to species level, whereas sedges, dicotyledonous herbs and small shrubs were categorised as forbs. A disc pasture meter (DPM) was used for comparing the standing biomass on kraal and control plots where the settling height of the disc was used as a measure of the grass biomass under the disc (Bransby and Tainton 1977 A PERMANOVA analysis (function adonis in the vegan package in R) was performed using Bray-Curtis distance and 999 permutations to determine whether kraaling had a significant effect on grass composition (where 'forbs' were included as a single 'taxon').

Statistics
The paired kraal and control sites were treated as replicates. Drivers of the response variables (vegetation and soil) were explored using a Pearson correlation matrix in R (ggcorrplot package 0.1.3, (R CoreTeam 2016)), taking into account both numerical and factor predictors.
Results of the correlation were used to guide the order of explanatory variables in subsequent linear mixed effects models (LME) and analyses of variance, after removal of autocorrelations including cation or anions in salts (e.g. Ca can be expected to autocorrelate with P and was not included in the P model). Predictors were kraaling and kraaling intensity (defined as large stock (450 kg) unit days, or LSU days), as well as site characteristics, such as site history (wattle infestation and fire), seasons since kraaling (1-5), slope, rainfall before and during kraaling, and site location (Table 1). Livestock included large stock (cattle) and small stock (sheep and goats) where small stock units were converted to LSU by multiplying with a conversion factor of 0.25. All data were inspected for normal distribution and assumptions of linearity before resorting to transformation as a corrective measure for bounded data (soil element percentages) or zero-inflated data (LSU days). Logit transformation was not necessary for sulphur (S), nitrogen (N) or soil organic matter (SOM), but improved data distribution in the case of calcium (Ca) and potassium (K). Soil phosphorus (P) remained non-normally distributed despite transformation. Poisson, quasi-Poisson and negative binomial regression models have been used more effectively than log transformations to redistribute zero-inflated data for linear models (O'Hara and Kotze 2010), but the LSU days data were not classic count data and did not fit any of these models (data not shown). Thus, LSU days data were cube root transformed and thereafter met all LME model assumptions of linearity. Analysis of variance was performed on response variables using ANOVA for normally distributed data that had been transformed where needed, or both ANOVA and the non-parametric Kruskal-Wallis test for soil P, which remained non-normally distributed. The importance of potential predictors of vegetation and soil change were tested using linear mixed-effects models (Harrison et al. 2018) with the 'lme4' package (Bates et al. 2015) in R (R Core Team 2016). Briefly, significance of the overall best-fit linear mixed-effects models used the maximum likelihood test and t-tests to determine the significance of terms within the model via the Satterthwaite's method, 'lmerModLmerTest', with an ANOVA-like table produced for random effects. Kraaling, kraaling intensity, slope, seasons or time since kraaling, and site history (wattle and fire history) were considered fixed effects, whereas site was considered a variable designating non-independence of replicates and was thus considered a random effect. All models met assumptions of linearity, including for soil P. Data in tables and figures are visualised as untransformed means.

Human and animal ethics
Data were collected and managed according to human ethics best practices where each participant was informed about the study, of their rights to anonymity and freedom to leave the study, after which they signed a standard consent form provided by the development organisation facilitating the study. Because researchers did not handle or manipulate the animals no animal ethics permit was required. All animals were managed by their owners as part of routine management, which normally included daytime herding and nighttime kraaling in mobile kraals located near stock posts. In discussion with livestock owners, the study considered all three 'Rs' of animal ethics (Knight 2012), namely replacement (we could not replace animals in this case), reduction (the least possible number of animals per herd were used), and refinement (supplemental licks were provided to animals in kraals that had no forage, to avoid weight losses as a result of the experiment).

Results
The main effect of overnight, short-duration kraaling was to increase soil P concentration by approximately 20% and S concentration by approximately 25% relative to non-kraaled sites (Figures 1a and 1b; p = 0.0426, df = 1, F = 4.691 for soil P; p = 0.0304, df = 1, F = 5.425 for soil S; two-way ANOVAs). Kraaling also marginally increased soil Ca from 0.1% ± 0.03 to 0.3% ± 0.11 (p = 0.0507, df = 1, F = 4.324). Increased kraaling intensity (LSU days or duration that livestock were in the kraal) was significantly correlated with increased soil P, S and Ca (r = 0.41, 0.43, 0.39; Supplementary Figure 1). Soil P and S changes in the study could be described by mixed effects models (Table 2), whereas the Ca model did not resolve. The most parsimonious models included kraaling treatment, kraaling intensity, and site as predictors (as well as site history for soil P) as follows: Soil P (%) = Kraaling treatment + Kraaling intensity + Site history + 1|Site, and Soil S (%) = Kraaling treatment + Kraaling intensity + 1|Site Here site refers to qualities inherent in the site and includes, but is not limited to site history of fire or wattle infestation (i.e. may include unmeasured site characteristics). The inclusion of site added to a low AIC value in the best model (Table 2). Kraaling did not increase soil N or SOM when measured one to three years after kraaling, but there was a positive relationship between soil N and year of kraal establishment, namely, kraals established most recently had significantly higher soil N (r = 0.54; Supplementary Figure 1) as one might expect. Short-duration kraaling did not increase heavy metals (arsenic, lead, cadmium, zinc, copper, mercury, nickel) or plant micronutrients, such as iron, zinc, manganese or molybdenum (data not shown).
A site history of fire or wattle infestation did not affect soil P or S (Figure 1; p > 0.05, two-way ANOVAs), but did result in a lower AIC for models. Previous wattle infestation increased soil Ca (0.40% ± 0.166), compared with soil where wattle had infested, but subsequently burnt (0.07% ± 0.007; p = 0.0426, Tukey test after a two-way ANOVA: p = 0.0433, df = 2, F = 3.689). Soil Ca concentration on unburnt grassland (0.15% ± 0.025) had intermediate values, but not significantly different from wattle infested sites with or without burning. The same trend was true for SOM, which decreased from ca. 11% in wattle infested areas to 8% after the wattle burnt, however here the effect was marginal (two-way ANOVA, p = 0.0623, df = 2, F = 3.200). As expected, fire presence was associated with decreased SOM and in our study, fire was always associated with wattle invasion (Supplementary Figure 1). Soil K was not affected by kraaling, but was increased by wattle infestation (without fire), slope and rainfall before and during kraaling (Supplementary Figure 1). Rain before and during kraaling determined the direction and magnitude of the effect of kraaling on infiltration where kraaling increased water infiltration when soils were wet from rain before and/ or during kraaling, but decreased infiltration when there was no rain during kraaling (correlation of LSU days and infiltration effect size, Pearson correlation coefficient = 0.66, p = 0.018, n = 12, data not shown). Kraaling over short time periods had no effect on soil bulk density (data not shown).   Table 2: Linear mixed effects models by maximum likelihood were used with Satterthwaite's method to provide ANOVA-like tables per term of the model explaining changes in soil P and S with kraaling or not in Eastern Cape, South Africa. In each case the AIC, BIC, chi square and p-values were compared between null (N) and test (T) models, and between all test models to select the most parsimonious overall model. The 'null' model always excluded the main fixed effect of interest (kraaling and kraaling intensity) and all models contained site as a random effect. Significance of the overall best-fit models and terms therein are indicated at p < 0.05*, p < 0.001** and p < 0.0001*** levels. The order of predictors in the models was based on relative importance in a correlation matrix (Supplementary Figure 1) Kraaling did not significantly change grass species composition (F (1,22) = 1.43, p = 0.19). The dominant plant species at most sites were Sporobolus africanus and Eragrostis plana. Other common species at both kraal and control sites were Aristida junciformis, Merxmuellera disticha (now Tenaxia disticha) and Heteropogon contortus, which were slightly more common on control plots, whereas Pennisetum clandestinum (now Cenchrus clandestinus) and forbs were more common at kraal sites. Site characteristics and their history also influenced vegetation characteristics. For example, sites previously infested with wattle (but not burnt) had reduced grass cover, compared with sites without wattle or fire (approximately 35% vs 8%, p < 0.05, a post hoc Tukey test after a one-way ANOVA with p = 0.0146, df = 2, F = 5.21). In addition, areas with greater slopes had more bare ground (more than 30% bare ground with slopes over 10%, R 2 = 0.53), whereas rainfall before or during kraaling had little influence on vegetation characteristics (Supplementary Figure 1).
The increase in soil P and S with kraaling occurred at most sites regardless of the initial condition, (i.e. the P and S concentrations of control sites) (Figures 2a and  2b). Soil organic matter was somewhat dependent on the initial condition with a small positive effect size, but this became negative when initial values were over 12% SOM (Figure 2c). In contrast, the effect of kraaling on vegetation cover was strongly dependent on the initial condition of that vegetation, which ranged between 5% and 100% herbaceous cover (Figure 3). Kraaling increased basal cover of grasses and forbs by ca. 50 and 15%, respectively, (and decreased bare ground by ca. 25%), but only where sites had very low initial basal cover, including areas where wattle had been cleared (Figures 3a, 3b and 3c). In contrast, intact grassland sites with initial basal and forb cover values of over 50% and 10%, respectively, showed a decline in cover by up to 15% with kraaling. The most variable and lowest average basal cover (44% ± 26.9%) was found at previously wattle infested sites that had not burned. Wattle sites that had burned and sites with no previous wattle infestation (which also had not burned) had relatively high initial basal cover (94% ± 3.0% and 96% ± 4.8%, respectively). Aboveground herbaceous biomass was decreased by kraaling generally, but especially when biomass was initially relatively high, >2 000 kg ha −1 (Figure 3d).

Discussion
Our findings highlight the importance of context of local vegetation type and evolutionary history when evaluating the efficacy of kraaling and other high intensity grazing practices. The type and intensity of historical herbivory differs between regions in relation to climate and soil nutrient status (Hempson et al. 2015) where only certain areas have conditions that support large herds and intensive grazing. The interaction between resource availability and evolutionary history of mammalian herbivory have led to differences in the life histories in the dominant grass species and hence how grass communities respond to concentrated grazing, trampling (Cingolani et al. 2005) and fire-herbivory interactions ( date, most studies on short-duration kraaling have been done in environments with an evolutionary history of high wildlife diversity and density, for example, Porensky and Veblen (2015) in Kenya, and Sibanda et al. (2016) and Huruba et al. (2018) in Zimbabwe, or sparse herbivory with transient large herds in the Karoo (McManus et al. 2018). In these environments, the effect of short-term kraaling generally increased soil nutrients, as well as palatability and basal cover of grasses.
Our study is the first to examine the effect of short-duration kraaling (7-24 d) in mesic montane grasslands and found that the effects of kraaling, despite the short duration, were much more variable than in higher nutrient environments, and potentially damaging to a vegetation dominated by bunch grasses. Results from previous studies conducted in other vegetation types thus needs to be extrapolated with caution when considering short-duration kraaling to manage mesic montane grasslands. Although soil P, S and Ca were increased only slightly in our study, P ions (and S in its elemental form) are relatively immobile in all but very sandy soils, and Ca is somewhat immobile (Marschner 1995), therefore these changes could be long-term changes. This may be detrimental to the long-lived perennial grasses that have adapted to high rainfall, leached and low nutrients soil and thus to the wild and domestic bulk-feeders that are adapted to feed on relatively low quality vegetation (Hempson et al. 2015). The effects of kraaling on N and SOM were small and/ or transient, likely because organic forms of N, such as urea from livestock easily volatilise as ammonia when on the soil surface, and inorganic N forms, such as nitrate and ammonium, are relatively mobile in soil and may be leached (Marschner 1995). Our other main finding was that many effects of kraaling on vegetation (and to a lesser extent soil) varied in magnitude and direction with the initial condition of the site. This was most obvious for grass basal cover where cover could be increased by kraaling only if the site was initially relatively bare, but at sites that had high basal cover beforehand (50% or higher), kraaling was damaging. Mesic grasslands are sensitive to heavy grazing and prone to soil erosion (Vetter 2006;Vetter et al. 2007). Long-term kraaling in mesic montane grasslands in South Africa resulted in more bare ground and a higher abundance of encroaching shrubs close to kraals (Morris 2017;Shezi et al. 2021), in contrast to the productive nutrient hotspots with nutritious, grazing tolerant grasses reported in parts of Kenya and Zimbabwe. Our results show that where there is still high basal cover, short-term kraaling should be avoided to maintain basal cover, and even though kraaling did not affect grass species composition in this short trial it is possible that with time even the more grazing adapted perennials (such as Sporobolus africanus, Eragrostis plana, Heteropogon contortus and Aristida junciformis) could be negatively affected by very heavy trampling.
In contrast, overnight and short-duration kraaling is promising as a tool for rehabilitating severely degraded mesic grassland sites. Apart from areas previously cleared of invasive woody plants, degraded areas in communal areas often include abandoned croplands that have lost topsoil and vegetation cover (Vetter 2007). Kraaling has been advocated as a way of fertilizing and restoring degraded or fallow croplands, and there is some evidence supporting this application (Peel and Stalmans 2018). Because kraaling is a valuable ancient practice to protect livestock from predators (Du Plessis et al. 2018), kraaling can be still conducted, but preferably on abandoned croplands or bare areas, as mentioned. Experimental kraaling at more sites with intermediate cover values would be useful to see whether a threshold (from increasing to decreasing effects of kraaling) can be identified. Our experiment was curtailed by drought, because carrying on would have been detrimental to livestock and possibly vegetation condition. It would be useful to resume the experiment to include a greater range of initial conditions and a more balanced design with respect to site history (fire and wattle).
Sites with a prior wattle infestation had lower basal cover and biomass, possibly due to microhabitat changes combined with allelopathic compounds from wattle (Souza-Alonso et al. 2013;Lorenzo et al. 2016). Effects of prior wattle infestation on soil characteristic appeared marginal unless combined with fire, which decreased soil Ca and especially SOM. Interestingly, some previously wattle cleared sites had high initial grass cover (those that were affected by fire) and they responded the same way to short-duration kraaling as the non-wattle sites. This suggests that besides kraaling, fire could be a potential way of restoring bare sites left after wattle clearing. As already mentioned, the impact of prior wattle infestation, with or without fire, on soil or vegetation characteristics was not an intended focus of the study and more intensive and balanced sampling would be required to explore effects.

Conclusions
In conclusion, we find that there is no one-size-fits-all solution regarding kraaling. The effect of short-duration overnight kraaling depends on the ecological context, as well as the initial condition of the site. The practical management recommendations based on our findings are that kraals (even if only present for a week) should not be positioned on intact, high altitude, mesic grasslands where basal cover is above 50%. Because kraaling is a valuable predator management tool and part of pastoralist livelihoods, we recommend that kraals in these grasslands are placed on abandoned cropland or otherwise bare areas, such as those left after clearing of invasive woody trees. Besides plant biomass and composition, rangeland managers should monitor and be aware of environmental factors, such as slope (steep areas are often relatively bare and may easily erode with kraaling) and timing of rainfall, because we found kraaling to decrease soil infiltration during non-rainfall periods.
Kraaling recommendations that are more broadly applicable to different vegetation types require that additional experimentation and monitoring take place in those areas. It is likely that in areas that have low nutrients and high rainfall or are otherwise resource limited, and that have historically relied on fire and relatively low densities of bulk grazers to cycle nutrients (Archibald and Hempson 2016), kraaling should be used conservatively. Our results show this is especially true for mesic grassland in good condition or with high basal cover, possibly because bunch grasses are not adapted to high animal densities and/or high nutrient deposition. In contrast, environments that support a variety and abundance of wildlife, and are less resource limited, may have grasses that are adapted to high animal densities and/or high nutrient deposition.
Finally, kraaling shows promise as a restoration tool to increase basal cover in degraded native grassland, for example, post-wattle clearing. Increases in soil nutrients due to kraaling also have potential for fertilizing croplands. Thus, apart from experiments in different areas, studies that include different initial conditions (such as wet vs dry soils) and intensity of kraaling (density and duration) would be very useful to further evaluate the potential utility of kraaling as a low-input restoration method, as well as a method to fertilise croplands prior to planting, or to rehabilitate those areas to grassland.

Data availability
Upon publication of this manuscript, the data that support the findings of the study will be openly available on ZivaHub Open Data UCT at https://doi.org/10. 25375/ uct.19096517.