A dominance shift in arid savanna: An herbaceous legume outcompetes local C4 grasses

Abstract The characteristic vegetation structure of arid savannas with a dominant layer of perennial grass is maintained by the putative competitive superiority of the C4 grasses. When this competitive balance is disturbed by weakening the grasses or favoring the recruitment of other species, trees, shrubs, single grass, or forb species can increase and initiate sudden dominance shifts. Such shifts involving woody species, often termed “shrub encroachment”, or the mass spreading of so‐called increaser species have been extensively researched, but studies on similar processes without obvious preceding disturbance are rare. In Namibia, the native herbaceous legume Crotalaria podocarpa has recently encroached parts of the escarpment region, seriously affecting the productivity of local fodder grasses. Here, we studied the interaction between seedlings of the legume and the dominant local fodder grass (Stipagrostis ciliata). We used a pot experiment to test seedling survival and to investigate the growth of Crotalaria in competition with Stipagrostis. Additional field observations were conducted to quantify the interactive effect. We found germination and growth of the legume seedlings to be facilitated by inactive (dead or dormant) grass tussocks and unhindered by active ones. Seedling survival was three times higher in inactive tussocks and Crotalaria grew taller. In the field, high densities of the legume had a clear negative effect on productivity of the grass. The C4 grass was unable to limit the recruitment and spread of the legume, and Crotalaria did outcompete the putative more competitive grass. Hence, the legume is able to spread and establish itself in large numbers and initiate a dominance shift in savannas, similar to shrub encroachment.

climatic conditions and hence they vary considerably. The majority of southern African savannas are semiarid or arid and receive less than 600 mm mean annual precipitation (MAP;Sankaran et al., 2005). Water is a limiting factor, and tree density is decreasing with lower MAP (D'Onofrio, Baudena, D'Andrea, Rietkerk, & Provenzale, 2015;Sankaran et al., 2005). Whereas at higher rainfalls trees are still abundant, the woody cover of the dryer savannas such as the thornbush savannas of Namibia is sparse.
There the vegetation is determined by a dominant layer of tufted perennial C 4 grasses in which trees and shrubs, mostly legumes, are sparsely interspersed. Annual grasses and forbs only occur during the rainy period in low abundances.
The characteristic vegetation structure in these water-and nitrogen-limited environments is regulated and adjusted by competitive processes, mainly between grasses and trees (Bond, 2008;Cramer, Van Cauter, & Bond, 2010;Donzelli, De Michele, & Scholes, 2013;Ward, Wiegand, & Getzin, 2013) but also other species (Sasaki & Lauenroth, 2011). Thereby, it is generally assumed that the living grasses use their superior competitive abilities (Cech, Edwards, & Olde Venterink, 2010) to maintain their dominance by suppressing the germination, reducing the growth, and thus regulating the establishment of trees and other competing species (Bond, 2008;D'Odorico, Okin, & Bestelmeyer, 2012;Sankaran, Ratnam, & Hanan, 2004). However, some studies contradict these findings and found, for example, seedlings of Acacia mellifera to grow and establish even when growing within grass tussocks (Joubert, Smit, & Hoffman, 2012;Rothauge, 2011). Furthermore, under water-and nutrient scarcity, facilitative interactions between grass tussocks and legumes gain in importance (Brooker et al., 2008;Maestre, Callaway, Valladares, & Lortie, 2009), as the tussocks store humidity and provide nutrients in form of litter (Maestre, Bautista, Cortina, & Bellot, 2001). Consequently, dead or dormant grass tussocks have even been found to support the germination and growth of other species seedlings (de Dios, Weltzin, Sun, Huxman, & Williams, 2014;Synodinos, Tietjen, & Jeltsch, 2015). These facilitative and competitive processes are modulated by the intensity and distribution of rainfall and the availability of water within the growing period.
Water availability does not only govern the germination of annuals and the rejuvenation of perennials but also influences the competitive balance between grasses and trees (Archer, Anderson, Predick, Schwinning, & Steidl, 2017;Joubert, Smit, & Hoffman, 2013;Kulmatiski & Beard, 2013;Woods, Archer, & Schwinning, 2014). If this well-balanced equilibrium is disturbed, sudden dominance shift can occur. Seedling of native shrubs or trees can gain the upper hand and lead to shrub encroachment (Sankaran et al., 2004;Scholes & Archer, 1997;Walter, 1953) or an "increaser" forb or grass can prevail and start a massive spreading (Smet & Ward, 2005;Vesk & Westoby, 2002). Both types of dominance shift in savannas are often associated with disturbances such as overgrazing or frequent fires.
In particular, shrub encroachment has become more frequent throughout the last decades (Higgins & Scheiter, 2012;O'Connor, Puttick, & Hoffman, 2014). Among disturbances such as land management changes and altered fire or grazing regimes, this increase is also attributed to higher atmospheric CO 2 -levels (du Toit & O'Connor, 2014), as elevated CO 2 concentrations decrease the photosynthetic disadvantage of C 3 plants or even favor them against the generally better performing C 4 grasses in hot environments (Bond & Midgley, 2000). In nutrient-limited savannas, increased CO 2 particularly favors legumes, as they are able to fix atmospheric nitrogen, strengthening their competitive abilities and affecting the competitive balance between grasses and legumes (Ward, Hoffman, & Collocott, 2014).
Although most dominance shifts in savannas known today are disturbance induced and the majority involve woody species (Eldridge et al., 2011;Sankaran et al., 2004;Scholes & Archer, 1997), there are a growing number of reports from farming communities in southern Africa about recent increases of native forbs obviously unrelated to disturbances. Likewise, the native herbaceous legume Crotalaria podocarpa has encroached parts of Namibia's escarpment region with considerable impact on fodder grass production and the lands carrying capacity (Wagner, Hane, Joubert, & Fischer, 2016).
With our study, we investigate the competitive balance between this legume and the dominant local fodder grass Stipagrostis ciliata and try to establish the causes for this recent spread of Crotalaria podocarpa. We carried out a controlled pot experiment to characterize the interaction between the C 4 grass Stipagrostis and the legume Crotalaria and supplemented this experiment with long-term field observations. We thereby hypothesize that: 1. Crotalaria is facilitated by Stipagrostis tussocks; 2. once Crotalaria is established, it has a negative effect on

Stipagrostis;
3. a negative effect on Stipagrostis is dependent on the Crotalaria density and independent of rainfall.

| Study area
The collection of soil and plant material for our interaction experiment and the field study took place on the farm Rooiklip, Khomas (S 23°24′23.29′′, E 016°03′37.35′′), situated 1,000 m a.s.l. in Namibia's lower great escarpment. Soils are predominantly shallow, nutrient-poor calcisols with a coarse texture and a high proportion of stones and sand (Wagner et al., 2016). The climate is hot-arid (Mendelsohn, Jarvis, Robertson, & Roberts, 2009). The scarce rainfall occurs predominantly between October and April with a pronounced precipitation maximum in February and March that defines the main growing season. Mean annual precipitation is 120 mm, while the mean precipitation during the growing season is 80 mm (Wagner and Fischer, unpublished data). The last decade (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014) was characterized by above-average rainfall and prolonged humid periods (Appendix, Figure S1), with a mean annual rainfall of 242 ± 57 mm and a mean February to March precipitation of 170 ± 42 mm. The vegetation is dominated by tufted perennial C 4 grasses of the genus Stipagrostis, mainly Stipagrostis ciliata (nomenclature according to De Winter, 1962; hereafter referred to as Stipagrostis). Together with the related S. uniplumis, it constitutes the main source of forage in Namibia's escarpment region (Juergens, Oldeland, Hachfeld, Erb, & Schultz, 2013;Müller, 2007). On the study site, Stipagrostis forms a light matrix of 1.5-2 tussocks/m 2 (Wagner et al., 2016) and its cover is considerably varying with seasonal rainfall. Loosely interspersed are perennial shrubs and occasionally trees. With sufficient rainfall, these perennials are complemented by annual grasses and forbs, including Crotalaria podocarpa. The area has not been used for farming or livestock-keeping for over two decades, but the area is used by free-roaming grazing wildlife such as zebra or gemsbok.

| Study species
Crotalaria podocarpa DC (hereafter referred to as Crotalaria) is an annual, herbaceous legume that is widespread in southern Africa (Polhill, 1968). It occurs on both sandy and stony soils, is well adapted to arid conditions, and has the capability to fix nitrogen (Jourand et al., 2005). Due to its content of pyrrolizidine alkaloids and flavonoids, Crotalaria is unpalatable to livestock (Wanjala & Majinda, 1999). The growth, number of flowers, and seed set of Crotalaria vary considerably with the amount of rainfall, and the plant produces a high number of seeds that are viable for more than 7 years.
Its comparatively large and heavy seeds are primarily dispersed by explosive dehiscence and reach a distance of about 5 m around the mother plant. Secondary long-distance dispersal is rare and probably related to scatter hoarding by ants and small mammals (Fischer, Kollmann, & Wagner, 2015). Crotalaria is part of the local plant community of Namibia's escarpment region where it normally occurs in moderate numbers. Starting in 2008, in the course of several years of above-average rainfall, a considerable proliferation of Crotalaria has been observed in Namibia's escarpment region (Wagner et al., 2016).

| Interaction experiment: competition and facilitation
To characterize the interaction between active and inactive (dead or dormant) Stipagrostis tussocks and Crotalaria seedlings, we carried out a greenhouse experiment under natural temperature regime and simulated natural rainfall conditions in March 2015.
Both inactive and active grass tussocks had approximately 5 cm basal diameter each and were clipped to 5 cm to ensure similar starting conditions and avoid shading. Three Stipagrostis tussocks each were planted in one pot (size 220 × 150 × 150 mm) to ensure intraspecific competition. Soil material and Stipagrostis tussocks were taken from an area of the farm that was unaffected by Crotalaria encroachment and free of Crotalaria seeds. Crotalaria seeds were collected in 2014 in Crotalaria-affected areas. Seeds were mechanically scarified with a scalpel and left for 24 h in petri dishes with 1 mm water to soak. Afterward, 10 swollen seeds with emerged radicle were evenly spaced 1 cm into the soil or the organic base of the grass tussock. Immediately after sowing, all pots were watered with a watering can equivalent to a rainfall event of 10 mm (equates to 330 ml/pot) for 3 subsequent days, and again after 5, 10, and 20 days. As Crotalaria typically reaches its flowering stage between 20 and 40 days (Wagner, unpublished we measured the number of surviving Crotalaria seedlings, maximum height of the emerging plants, and the maximum length of Stipagrostis culms for 34 consecutive days. Seedlings, which did not yet have cotyledons, were wilting, or damaged were excluded from the measurement. Complementary to this, we characterized the water retention capacity of the respective substrates bare soil, soil with inactive grass tussocks, and soil with active tussocks using time-domain reflectometry (Ledieu, De Ridder, De Clerck, & Dautrebande, 1986;Robinson, Jones, Wraith, Or, & Friedman, 2002). Each pot was measured at 7 hr in the morning with three repeats over 10 days after a simulated rain event of 10 mm.

| Field observations: Stipagrostis productivity and density
We used complementary field data to characterize the interaction between Crotalaria and Stipagrostis tussocks. Our field observation was carried out on 20 long-term observation plots of 10 × 10 m half of which were affected by Crotalaria encroachment (Wagner et al., 2016). All plots had identical soil types, similar soil texture, and initially the same vegetation composition and structure (Table S1)

| Data analysis
Statistical analysis was performed with R 3.2.1 (R Core Team, 2015).  Figure S2). We further quantified the effect of Crotalaria density on the productivity of Stipagrostis tussocks in relation to rainfall (average tussock area/seasonal rainfall) using a linear mixed-effects model with plot as a random factor to ensure independence of errors with respect to temporal autocorrelations (Pinheiro & Bates, 2000). Rainfall lower than 40 mm was excluded in this model, as here the area covered by grass tussocks was reduced to their basal area and due to a high frequency of outliers. To obtain normality of variances, response, and explanatory, variables were log-transformed.

| Results of the interaction experiment
In our interaction experiment, seedling survival and growth of Crotalaria were clearly facilitated by inactive Stipagrostis tussocks.
Grass tussocks were able to maintain soil humidity longer than bare soil. The moisture content of bare soil dropped below the permanent wilting point of coarse sand (~2%) three days after the rain event. Active tussocks maintained a moisture level above the permanent wilting point for 5 days and inactive tussocks even for 6 days ( Figure   S3). All grass tussocks maintained their initial activity state during the whole experiment.
Time and maximum number of germinated Crotalaria seeds that did reach a measurable seedling stage (at least cotyledons developed, size >20 mm) varied considerably between the treatments.
From 150 seeds planted on bare soil, only 40 reached this stage after 10 days. Seeds planted in Stipagrostis tussocks reached this stage 2 days earlier and survived in higher numbers, with 70 seedlings in active tussocks and even 109 seedlings in inactive tussocks (Table   S2). Overall seedling survival was three times higher when growing in inactive tussocks (mean ± SE: 62.7 ± 8.1%) than when growing alone (18.7 ± 6.0%) or in active Stipagrostis tussocks (19.3 ± 7.6%).
Stipagrostis showed no growth for the first 2 days. Between Day 3 and Day 7, Stipagrostis growing without Crotalaria and Stipagrostis under competition with Crotalaria both had the same high growth rate of 1.55 ± 0.26 cm/day, but from Day 8 on, growth rates slowed considerably and exhibited significant differences between Stipagrostis without competition with 0.28 ± 0.05 cm/day and Stipagrostis under Crotalaria competition with 0.16 ± 0.03 cm/day (Figure 1b, Table S4).

| Results of the field observations
During the whole study, only moderate grazing through zebra occurred, but no exceptional herbivory event, for example, through arthropods, was recorded. The number of Crotalaria seedlings growing within Stipagrostis tussocks was disproportionally high compared to the number of seedlings growing outside grass tussocks: On average, 60.0 ± 1.7% of all Crotalaria seedlings were found to grow within grass tussocks, which covered only 3% of the plot area. The Neighbor-Effect Intensity Index NInt A of Crotalaria on Stipagrostis was consistently negative over the whole study period, indicating a pronounced competitive effect of Crotalaria on Stipagrostis (Table 1).

| D ISCUSS I ON
We found neutral or even facilitative effects of Stipagrostis on Crotalaria. Inactive (dead or dormant) grass tussocks clearly increased growth and survival of Crotalaria seedlings, whereas active grass tussocks did not affect seedling survival and growth of Crotalaria. Contrastingly, the effect of Crotalaria on Stipagrostis was consistently negative and increased with Crotalaria density, independent of rainfall.
Our interaction experiment covered the crucial early stages that are decisive for the successful recruitment of the legume (Bond, 2008;Harper, 1977;Wiegand, Saltz, & Ward, 2006). The interaction between Stipagrostis and Crotalaria was both facilitative and competitive, but clearly to the benefit of Crotalaria and to the detriment of the grass. During the growth of the legume, the interaction between Crotalaria and Stipagrostis undergoes a shift, and, while seedlings are facilitated by the grass, the legume later outcompetes or at least restricts (growth reduction) its former nurse plant. Similar ontogenetic shifts under arid conditions involving a grass as nurse plant have only been described for shrubs, such as Stipa tenacissima and Lepidium subulatum (Soliveres, Desoto, Maestre, & Olano, 2010) or Agrostis magellanica and Azorella selago (Roux, Shaw, & Chown, TA B L E 1 Rainfall, Stipagrostis tussock area and Neighbor-Effect Intensity Index NInt A of Crotalaria on Stipagrostis productivity (tussock area) between 2009 and 2015 based on the average individual cover of tussocks on affected and unaffected field sites  (February, Higgins, Bond, & Swemmer, 2013), Crotalaria significantly reduced the growth rate of Stipagrostis. This negative effect of Crotalaria on Stipagrostis extends over the later growth stages of Crotalaria, increases with higher legume density, and is, other than the recruitment of Crotalaria, not compensated for by higher rainfalls and water availability. Thereby, our interaction experiments support the findings of Maestre, Bautista, and Cortina (2003) who found seedlings of legumes to grow largely unaffected in living grass tussocks under stressful conditions similar to those of our study region. In the case of Crotalaria, the C 4 grass Stipagrostis is not, or probably no longer able to exert its supposed dominance (February et al., 2013;Riginos, 2009;Sankaran et al., 2004;Scholes & Archer, 1997) over the legume and suppress its recruitment. This clearly contradicts the widespread opinion that grasses are able to outcompete legumes and encroachers due to their competitive advantage in water-limited environments Riginos, 2009).

This raises the question why such a dominance shift involving
Crotalaria has not been observed before. As classic disturbances such as increased grazing, herbivory events, or fire can be ruled out, we propose two possible explanations: Either the competitive capabilities of Crotalaria have recently gained in strength due to more favorable conditions or they are still unchanged but the incidence of competition has changed. In the latter case, the elevated rainfalls between 2007 and 2011 ( Figure S1) might be a trigger. Higher rainfall is associated with higher seed production and generally better seedling survival of encroaching species (Kraaij & Ward, 2006;Oldeland, Dreber, & Wesuls, 2010;Roques, O'Connor, & Watkinson, 2001). Taking into account Crotalarias high seed production, its extended seed viability, and dispersal by explosive dehiscence (Fischer et al., 2015), above-average rainfall will automatically be linked to a higher probability and number of seeds ending up in grass tussocks after dispersal. Our pot experiment has shown that the grass tussocks facilitate early establishment of Crotalaria and that the grass is unable to outcompete 3-4 Crotalaria seedlings per tussock. Our field data clearly demonstrate that the negative effect of Crotalaria on Stipagrostis increases with Crotalaria density. Consequently, there is now not only a higher probability of grass tussocks being faced with competing Crotalaria, but the affected tussocks may also have to compete with a higher number of Crotalarias, as more seedlings survive. This effect is further intensified as the perennial grass tussocks may already be weakened by competition during the preceding season.
Regarding a strengthening of Crotalarias competitive abilities, we may also speculate about a possible role of elevated atmospheric CO 2 levels. Increased atmospheric CO 2 has been recently identified to be an important determinant of encroachment processes (Higgins & Scheiter, 2012;Kulmatiski & Beard, 2013;O'Connor et al., 2014;Ward et al., 2014) as it favors C 3 over C 4 plants. Possible nutrient deficiencies can be compensated for by nitrogen fixation, provided temperatures are not too high, and there is sufficient water available (Sita et al., 2017). Once the seedlings have survived the early stages, they might then be able to grow more quickly and eventually overgrow the grasses. Already impaired by competing with a number of Crotalaria seedlings, the light-sensitive grass tussocks (Zimmermann, Higgins, Grimm, Hoffmann, & Linstädter, 2010) later are additionally affected by shading and might be weakened even more, further contributing to a vicious circle at the end of which the dominance shift occurs.

| CON CLUS ION
The native herbaceous legume Crotalaria podocarpa does not only withstand the competition from the perennial C 4 grass Stipagrostis ciliata but also has a pronounced negative effect on the grass. Likely due to the high number of seeds produced after higher rainfall, Crotalaria is able to prevail over the grass and initiate a dominance shift in the arid savanna of Namibia's escarpment region. The ramifications of this shift resemble those of shrub encroachment or invasion, but it involves a native and herbaceous legume and is obviously not triggered by disturbance.

ACK N OWLED G M ENTS
We are grateful to numerous students for their help with the fieldwork and H. Neuffer and F. van Biljon for their logistic support and for allowing us to carry out the study on their farm. We thank our reviewers for their valuable comments and Sonja Bernhard for language check. Our research was approved by the Ministry of Environment and Tourism of Namibia, research permit 1982/2014.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
TCW, CF, and DJ conceived and designed the experiments. JR and TCW performed the experiments and collected the data. CF and TCW analyzed the data. TCW and CF wrote the manuscript. All authors contributed critically to the draft and gave approval for publication.

DATA ACCE SS I B I LIT Y
Data will be made available through OSF.