Plant community structure and possible vegetation changes after drought on a granite catena in the Kruger National Park , South Africa

The Earth’s environment is dominated by three great natural components, namely, climate, vegetation and soil. Climate is considered the most important factor influencing the distribution and composition of vegetation on a micro and sub-continental scale (Campbell et al. 2008; Furley 2010; Scholes 1997; Schulze 1997). Vegetation development is controlled largely by light, temperature and moisture (Bond, Midgley & Woorward 2003; Schulze 1997). Topography and the chemical and physical compositions of the soil also influence vegetation and, in conjunction with climate, are responsible for the intricate interactions that govern the worldwide distribution of vegetation (Campbell et al. 2008; Furley 2010; Scholes 1997). Understanding how these interactions regulate the ecology of plant communities is critical for characterising the impacts of global change on biodiversity at local and regional scales.


Introduction
The Earth's environment is dominated by three great natural components, namely, climate, vegetation and soil. Climate is considered the most important factor influencing the distribution and composition of vegetation on a micro and sub-continental scale (Campbell et al. 2008;Furley 2010;Scholes 1997;Schulze 1997). Vegetation development is controlled largely by light, temperature and moisture (Bond, Midgley & Woorward 2003;Schulze 1997). Topography and the chemical and physical compositions of the soil also influence vegetation and, in conjunction with climate, are responsible for the intricate interactions that govern the worldwide distribution of vegetation (Campbell et al. 2008;Furley 2010;Scholes 1997). Understanding how these interactions regulate the ecology of plant communities is critical for characterising the impacts of global change on biodiversity at local and regional scales.
The savanna biome is unique because it consists of both woody vegetation and a grass layer. Climate and other regulating factors likely affect these two components differently, resulting in spatio-temporal heterogeneity of tree:grass compositions. Severe droughts, for example, may remove trees, leading to negative effects on woody plant diversity (Swemmer 2016;Walker et al. 1987;Zambatis & Biggs 1995). By reducing tree densities, droughts in savanna provide opportunities for drought-adapted flora to thrive, for instance, by promoting seedling recruitment of fast-growing, palatable shrub species and the re-establishment of a grassy layer (Swemmer et al. 2018;Vetter 2009). In this way, drought can help maintain the balance between trees and grasses (Swemmer 2016). Grasses, on the other hand, can take decades to recover their A preliminary study investigated the associations between vegetation communities along catenary soil gradients in 2015. The severe drought of 2016 in South Africa presented the opportunity to study post-drought savanna vegetation changes. This hillslope transect was surveyed for five successive seasons. The Braun-Blanquet method was used, and the data were analysed by means of the TWINSPAN algorithm, which resulted in the classification of different communities on the crest, sodic site and riparian area. Change in herbaceous and grassy vegetation composition and diversity in the transect is compared between rainfall years, wet and dry seasons, and three different zones (crest, sodic site and riparian areas). Spatial and temporal autocorrelation of the woody component shifted the focus to variance within the graminoid and herbaceous layers. Clear vegetation changes were observed on the crest and the sodic sites, whereas changes in the riparian area were less obvious. In all three habitats, species richness decreased after the drought and did not reach pre-drought levels even after two years. However, plant species diversity was maintained as climax species were replaced by pioneer and sub-climax species. These changes in community structure, which had reverted to systems dominated by climax species by the end of the sampling period, might be an indication of the savanna ecosystem's resilience to drought conditions. http://www.koedoe.co.za Open Access productive potential or might recover comfortably before the next drought (Swemmer et al. 2018). The herbaceous layer thus also regularly experiences negative responses to drought (Zambatis & Biggs 1995); however, Abbas, Bond and Midgley (2019) indicated that grasses can resprout vigorously after the onset of rainfall events. In fact, this layer usually responds to droughts and other climate changes first, primarily because of the shallow depth of root penetration. Upper soil layers are more susceptible to desiccation than the deeper strata penetrated by many woody plants. Furthermore, the extensive root structures of trees increase their access to subterranean reserves of ground water. Shorter term responses of grassy and herbaceous vegetation were highlighted by Buitenwerf, Swemmer and Peel (2011), who showed that dynamics of this savanna component are mainly controlled by interannual changes in rainfall. The response of the grass layer to climate is of importance for conservation planning and application, because it is an important food source for grazer populations (Staver, Wigley-Coetzee & Botha 2018).
The savanna regions of South Africa are considered semiarid, receiving rainfall mostly during the summer months between October and April (Walker et al. 1987). Fluctuations in annual rainfall, including droughts, are a regular and recurrent feature of the climate .
In more than half of the 80 summer rainfall districts identified by Rouault and Richard (2003), droughts were recorded during 1926during , 1933during , 1945during , 1949during , 1952during 1970during , 1983during and 1992during (Fauchereau et al. 2003Gommes & Petrassi 1996). Rouault and Richard (2003) and Staver et al. (2018) indicated that the 1982-1983 drought was the worst drought recorded since 1922; however, Swemmer (2016) indicated that the drought of 2015-2016 was the worst drought that the Lowveld experienced in the past 33 years. In the savanna areas of KwaZulu-Natal, this drought was shown to be the worst in 50 years by Abbas et al. (2019). Research by Hu and Fedorov (2019) indicated that the drought of 2015-2016 was worse than the droughts of 1982 and 1997. These studies show that, since the 1960s, drought is more often associated with El Niño events; notably, however, annual rainfall during wet years has also increased since the 1970s.
South African savannas experienced drought conditions during the rainfall seasons of 2014-2015 and 2015-2016. In the Kruger National Park (KNP), and the surrounding areas of the Lowveld, below average rainfall occurred at annual (255 mm) and monthly scales (Swemmer 2016). This resulted in devastating effects on vegetation, animal and human welfare in certain areas. These years were also marked by unusually high temperatures, resulting in higher evaporation rates, further reducing water availability (Swemmer 2016). The severity of these conditions provided us with the opportunity to study their effects on short-term responses of vegetation, specifically on the grassy and herbaceous component. We conducted a study of seasonal and annual plant community dynamics along a granitic catenal gradient. This catena forms part of a research supersite, where long-term research is needed to establish baselines for monitoring and understanding ecological change (Smit et al. 2013). We describe taxonomic community changes, as well as testing for shifts in diversity, over two wet and two dry seasons through the drought period and compare these with pre-drought conditions (April 2015) described elsewhere (Theron, Van Aardt & Du Preez 2020). We focused only on the herbaceous and grassy components of the vegetation because we were interested in resolving short-term responses in savanna plant resilience to drought.

Study area
The study site is in the southern parts of KNP south of Skukuza (see study area figure in Theron et al. 2020) at 25.111ºS and 31.579ºE. Kruger National Park falls within the arid 'BSh' (hot semi-arid climate) climate type according to the Köppen-Geiger classification system (Venter, Scholes & Eckhardt 2003). 'BSh' is one of the four climate types within this category. The main features of 'BSh' climate are distinct seasonal rainfall and temperature variations. Mean annual precipitation in KNP is generally in the range of 650 mm annually (Smit et al. 2013). On a local scale, MAP of the Granite Lowveld varies between 450 and 900 mm along the eastern plains and the western escarpment, respectively (eds. Mucina & Rutherford 2006). However, the average annual total rainfall as recorded at the Skukuza Meteorological Station is 553 mm (Zambatis 2006). The mean annual temperature in the vicinity of the study area varies between 21ºC and 22ºC (Khomo et al. 2011;Scholes, Bond & Eckhardt 2003). This area experiences an insignificant seasonal and diurnal temperature variation with extreme periods of inundation and aridity (Kruger, Makamo & Shongwe 2002). The study site is underlain by the Nelspruit Suite geological formation and consists of granite and gneiss mostly occurring in the eastern parts of KNP (Alard 2009;Smit et al. 2013;Van Zijl & Le Roux 2014). Granite gneiss is widespread in the eastern regions of KNP and results in shallow, nutrient-poor soils that vary from grey to red to brown in colour (Venter 1990). Descriptions of the different soil forms found along the catena at the site were provided in Figure 2 within the article by Theron et al. (2020). The vegetation type at the study site is mostly Granite Lowveld (SVI3), characterised by a ground layer of tall grasses with

Data collection
The same hillslope transect was surveyed for five seasons; the first survey was conducted prior to the onset of severe drought conditions (Theron et al. 2020) during December 2015 and April 2016 (Figure 1). The second and fourth surveys represent the start of the rainy summer season, while the third and fifth surveys reflect the end thereof ( Figure 1). Relevés of 10 m 2 were aligned along a 500 m transect. Cover abundance was recorded per species according to the modified Braun-Blanquet scale (Kent 2012;Kent & Coker 1992;Van der Maarel & Franklin 2013;Theron et al. 2020).

Classification, richness and diversity analysis
The analysis done by Theron et al. (2020)

Classification
VegCap (unpublished database tool designed by N. Collins) was used to capture vegetation data into a macro-enabled Excel spreadsheet. From there, the data were imported into JUICE© (Tichý & Holt 2006) where a Modified TWINSPAN Classification (Roleček et al. 2009) analysis was carried out. Parameters for this analysis included the following: pseudospecies cut level (5); analysis was constrained to a minimum group size of 3-54 clusters; and division reached an endpoint if dissimilarity went lower than 0.3 based on average Sorensen dissimilarity. The resulting clusters were then arranged within both JUICE© and Excel to form the final vegetation communities. Although all the species were recorded during the field surveys, woody species were removed from the data in order to look at the change in graminoids and herbaceous species after the drought. This follows, for example, Rouault and Richard (2003), who indicated that trees and other vegetation with extensive root structures have access to subterranean reserves of groundwater and will thus not be immediately affected by the drought. The naming of communities and subcommunities was carried out according to the guidelines presented in Brown et al. (2013). In order to obtain diagnostic, constant and dominant species, we made use of the Analysis of Columns of a Synoptic

Diversity and richness
In addition to descriptions of community composition and how this changed over time, we evaluated changes in diversity and compared these across time for each of the three communities. We compared changes in species richness as well as changes in alpha-diversity. We used the Chao estimator as an indicator of species richness, as this index accounts for the occurrences of singletons and doubletons, and the Shannon index was used to quantify alpha-diversity. For each sample (i.e. per season and per habitat), ordinal abundance data as scored by the Braun-Blanquet system were converted to abundance cover data, rounded to integer values, following Van der Maarel (2007): r = 1; + = 2; 1 = 3; 2a = 8; 2b = 18; 3 = 38; 4 = 63; 5 = 88. Diversity estimates were computed using the iNext package (Hsieh, Ma & Chao 2016) for R (R Core Team 2015). The iNext function was used for extrapolation and prediction of diversity indices based on rarefaction procedures, with the expected means and standard errors extrapolated from the asymptotes of the fitted accumulation curves (see Figure 2). In all cases, accumulation curves approached or reached an asymptote, and observed data represented between 80 and 100% of extrapolated estimates (in the case of species richness), and between 94% and 100% of extrapolated estimates (for Shannon diversity), depending on the sample. Thus, sampling effort is considered sufficient for reliable estimations of diversity in these communities.

Ethical considerations
Ethical approval was obtained from the Interfaculty Animal Ethics Committee of the University of the Free State (UFS-AED2019/0121).

Results and discussion
Classification Different plant communities were classified for each topographical unit as defined by Theron et al. (2020). In this article, the data for 2015 were not included in the classification in order to prevent a repetition of information.

Crest communities (December 2016-April 2018)
These communities located on the crest zone and upslope beyond the sodic site occur on the Clovelly, Pinedene, Fernwood, Estcourt, Mispha and Sterkspruit soil forms (Theron et al. 2020). The soil depth varies from 533 to 620 mm deep, with an average pH H₂O of 5.95-6.08. Soil texture is mostly loamy sand to coarse loamy sand (Theron et al. 2020).

Sodic site communities (December 2016-April 2018)
The communities occur between the crest and the riparian area on the mid-slope of the hill, and are also sodic sites. Soils are mostly of the Sterkspruit form; however, there were also instances of Mispah soil forms present. The depth varies between 180 mm and 500 mm with an average pH H2O of 6.20-6.43. Soil texture is coarse sandy loam. The vegetation classification resulted in two communities and four sub-communities (Online Appendix 2). In terms of vegetation composition, these communities can be compared to the Dactyloctenium aegyptium-Sporobolus nitens (community 4) of Theron et al. (2020) The vegetation found in this sub-community mostly represents species from growing season 4 with a single occurrence of season 2. Species from Species Group A (Online Appendix 2) define this sub-community. These species are completely absent or occur with very low cover-abundance values in other communities and sub-community on the sodic site. Urochloa panicoides, which defines this sub-community, is known as a pioneer annual tufted grass and will thus only be present for one season (Van Oudtshoorn 2018). In this subcommunity, this grass co-occurs with Sporobolus nitens, which defined the communities found in 2015 before the drought: Vegetation in this sub-community is mostly from growing season 5 (April 2018) with a single occurrence of vegetation from growing season 3 (April 2017). Although S. nitens is the diagnostic species for this sub-community, the presence of species from Species Group D defines this sub-community. These species are completely absent from sub-community 2.2. Season 5 marks the return of S. nitens (with high cover abundance) and Dactyloctenium aegyptium (with low cover-abundance and only in some relevés) which dominated the communities found on the sodic site by Theron et al. (2020) in 2015:

Chloris virgata-Eragrostis cylindriflora-Chloris gayana
Sub-community Sub-community 2.2 is defined by the presence of perennial grasses from Species Group E, which are absent from subcommunity 2.1. Although having low cover abundances and not occurring in all relevés, this is the only season in which these grass species were found. All three of these grass species (Chloris gayana, Eragrostis gummiflua and Aristida stipitata) are regarded by Van Oudtshoorn (2018) as sub-climax species, which might indicate that after the third season, the sodic site started to recover from the severe drought of 2015-2016.

Riparian area communities (December 2016-April 2018)
The communities occur between the sodic site on the lower midslope of the hill and the drainage line. Soil forms found in this area include Dundee, Mispah, Bonheim and Sterkspruit. The depth of these soils varies from 100 mm to 600 mm with an average pH H20 of between 6.21-6.73. Soil texture also varies from sandy loam to loamy to sandy clay loam. In contrast to the other terrain units depicted along the catena, the riparian area's classification did not result in communities that could depict the different seasons of sampling. The vegetation classification resulted in five communities (Online Appendix 3). The vegetation of the riparian communities can be compared to communities 1 (Panicum maximum-Pupalia lappacea) and 2 (Themeda triandra-Flueggea virosa) from Theron et al.'s (2020)  Eragrostis cylindriflora (Species Group G) and Urochloa mosambicensis (Species Group H) define this community. Species from Species Group A are mostly present in community 1 and absent or occur with low cover-abundance value in other communities in the riparian areas. This community represents sampling seasons 2, 4 and 5. It is notable that none of the relevés done during season 2 (just at the onset of the rainy season) is present in this community. Community 1 also share a lot of species from Species Group B with community 2:

Themeda triandra-Panicum maximum Community
Diagnostic species: None Community 2 is defined by the presence of species from Species Group C, which are mostly restricted to this community although they occur with low cover-abundance values. Notable in this community is the strong presence of Themeda triandra (Species Group D) and Panicum maximum (Species Group H), which were also present as diagnostic species defining the riparian areas in Theron et al. (2020). It seems as if Themeda triandra is mostly limited to this community with high cover-abundance values. However, Panicum maximum occurs throughout all the communities present in the riparian area throughout all the sampling seasons. This community is also mostly represented by sampling seasons 3 and 5 with some instances of sampling season 4: Community 3 is the community with the lowest number of species in all the communities found in the riparian area, and there are no species that clearly distinguish this community from all the other communities in the riparian area. The cover abundance of species in this community is also low, and species do not occur in all the relevés found in this community. It is only the grass Eragrostis superba (Species Group H), known to grow in disturbed areas (Van Oudtshoorn 2018), that occurs in all three relevés that make up the community. Vegetation in this community mostly represents sampling seasons 2 and 5. The reason for the low number of species might be that the vegetation still needed to recover after the drought.

Eragrostis rigidior-Urochloa mosambicensis Community
Diagnostic species: None Vegetation in this community is dominated by species from Species Group E, which are mostly absent from the other communities in the riparian area. Furthermore, Urochloa mosambicensis (Species Group H) also occurs more frequently and with a higher cover abundance in this community. According to Van Oudtshoorn (2018), U. mosambicensis grows in disturbed or overgrazed and trampled areas. The high occurrence of this species in the riparian area might indicate that animals were seeking shade in order to evade the heat of the day during the drought (2015)(2016). He further also indicated that Eragrostis rigidior is known to occur in disturbed soil. It is also important to note that most of the relevés present in this community represent sampling season 2, which was just after the 2015-2016 drought: This is the only community that is solely represented by vegetation sampled during sampling season 4. The vegetation is mostly dominated by the presence of species from Species Group F, which is mostly absent or occurs with low coverabundance values in other communities of the riparian area. The grasses Bothriochloa radicans and Eragrostis trichophora are known to occur in areas with additional moisture or where water collects (Van Oudtshoorn 2018). A possible explanation for this might be that after rains, water can remain close to the surface in the vicinity of the riparian area, which contributes to the additional moisture that is favourable for these grasses.

Bothriochloa radicans-Eragrostis superba Community
Although there is no distinction to be made between the sampling seasons in the riparian area of the study site, there are differences in the vegetation composition over the study period. When comparing the vegetation of the riparian area with communities 1 and 2 (Theron et al. 2020), it is clear that Panicum maximum, Urochloa mosambicenis and Themeda triandra remained an important part of the vegetation composition over all the different sampling seasons.

Richness and diversity of plant communities
From Figure 3a, it seems as if the species richness decreased at all the sites during the drought and subsequently increased more-or-less progressively through time as the communities recovered from the drought between 2015 and the onset of the current sampling period. However, pre-versus post-drought richness estimates are only significantly different for the sodic and riparian habitats (non-overlapping 95% confidence intervals between groups); variance in estimates for the crest communities is high and overlaps with the pre-drought estimate. Interestingly, however, the recovery in species richness in http://www.koedoe.co.za Open Access sodic and riparian habitats appeared to slow or even reverse by the end of the study period (April 2018), although this could be because the final sample was taken in the dry season. Overall, species richness in crest habitats was greater than in both sodic and riparian habitats. Figure 3b represents the changes that took place in diversity over the different sampling seasons. In contrast to richness, species diversity did not differ between pre-and post-drought periods. However, a more cyclic seasonal shift is apparent, in that diversity was often highest in the wet seasons (December samples), compared with both dry season samples (April). The sodic and riparian habitats are an exception to this trend, because diversity in these areas was low in December 2016, perhaps because of a lag in recovery from the drought. As with species richness, diversity was also consistently greater in the crest, compared with the other two habitats.
While these indices of diversity provide some indication about changes in the studied communities, their overall function might be better represented in terms of changes in plant functional groups. Indeed, in all three habitats, the proportional representation of plant functional groups differed between 2015 and 2016, with climax and subclimax species being replaced by pioneers, perennials, annuals and -in some cases, especially in the sodic habitat -bare soil ( Figure 4). By the end of the sampling period, however, the frequency distribution of functional groups at each habitat was qualitatively similar to pre-drought conditions.

General discussion
With this study, we aimed to determine how savanna plant communities along a catenal gradient changed over time following a severe drought. The catenal gradient studied could be divided into three plant communities -crest and midslope with the highest diversity; sodic site, and riparian areas. The crest and sodic sites further showed a definite change in species composition among the different sampling seasons. There was also an association between April sampling seasons for the crest as well as associations between the December and April sampling sites for the sodic site. Vegetation in the riparian section of the study revealed no clear distinction between different sampling seasons or any correlation between April and December. In a study by Scholes (1985), he investigated the drought of 1981-1983 and found that the grasses were more adversely affected by the drought than the trees. Although we excluded data for woody plants from this study, it is clear that vegetation changes took place in the ground layer (graminoids, forbs, herbs and geophytes), especially in the crest and sodic site communities (see Janecke 2020).
Previous studies have indicated that the physical and chemical properties of soils would affect grass mortality rates during drought conditions (Khomo & Rogers 2005;Khomo et al. 2011;Scholes 1985). Specifically referring to the characteristics of the study site and its catenary properties, it is expected that grasses inhabiting the sandy crest and valley bottoms would have a higher mortality rate than those inhabiting the clay-rich sodic sites and downslopes. The physical properties of sandy soils would compound the effects of droughts because they retain less water than do clay soils, and also through exasperating water infiltration and percolation of any available surface water. The effect of soil properties was shown to also affect this catena complex (Theron et al. 2020). This is also comparable to this study because most of the grass species dominating the climax community (sampling S1; 2015) returned to the vegetation composition of communities during sampling season 3 (April 2017). We furthermore found that richness and diversity declined and that recovery was not complete two years after the drought, especially in the sodic and riparian habitats, which have maintained a low level of species richness throughout the sampling period. These shifts coincided with changes in functional group representation following the drought.

Conclusion
Definite changes in plant community composition were seen in the crest, midslope and sodic sites during the different sampling seasons. Shifts were also seen in terms of species composition at certain times of the year. This was not always clear in terms of richness and diversity of plant species. We would, however, be cautious to extrapolate these findings to all vegetation successions along a catena.
In the riparian area, no distinctions were clear between the different sampling seasons and no cyclic correspondence was observed between April and December. This phenomenon might be ascribed to water movement through the process of hydraulic lift from deeper soil layers which lessen the impact of drought on the vegetation.
We recommend that future studies following droughts should be done over more sampling seasons than reported here to better relate seasons to plant assemblages. Lastly, the recovery of the plant growth forms from 2015 to 2018 might be an indication of the resilience of the savanna ecosystem, in spite of the recovery not being complete.