Ecological stability–the tendency of aggregate community properties such as primary productivity to be stable through time–is a key metric for understanding ecosystem dynamics and predicting responses to a changing environment. Stability can be shaped by bottom-up processes, for instance when ecosystem properties like primary productivity and total vegetative cover directly track environmental resource fluctuations (Power, 1992; La Pierre et al., 2011; Craine et al., 2012). At the same time, stability can be structured by top-down processes, such as when episodic herbivory temporally shifts plant investment in productivity, or when the effect of herbivory shifts seasonally (Hairston et al., 1960; Van Langevelde et al., 2003; Porensky et al., 2013). However, top-down processes remain understudied relative to environmental variability. Finally, stability can be influenced by dynamics among species within the community. Biotic stabilizing mechanisms occur when species dynamics result in a community that is more stable than predicted by external variability. The presence of such dynamics complicates assessment of the impacts of bottom-up and top-down drivers; beyond their direct impacts on the stability of ecosystem properties, we must also consider indirect effects via influences on species composition and community dynamics.

Multiple biotic stability mechanisms can contribute additively or differentially toward the stability of aggregate community properties (Lehman and Tilman, 2000). These include: (a) asynchronous dynamics, also known as compensatory dynamics, where declines in one species are compensated for by a rise in another, (b) stable dominant species, where aggregate stability depends on a few dominant species with high population stability, or (c) species richness driving portfolio effects, in which the community property is distributed among multiple species in a diverse community (especially when asynchronous). A growing body of work has highlighted that biotic stability mechanisms depend on the environmental context, with most studies exploring this in relation to bottom-up resource dynamics (Grman et al., 2010; Hallett et al., 2014; Xu et al., 2015; Hautier et al., 2020). Biotic stability mechanisms can shift along environmental gradients (Hallett et al., 2014), and resource variability can also cause shifts in mechanisms over time in the same system. For example, drought can both reduce biomass (Craine et al., 2012; Aroca, 2013; Grant et al., 2014; Alon and Sternberg, 2019) and increase variability if production tracks rainfall fluctuations (La Pierre et al., 2011). Both a decline in the mean and increase in the (relative) variance can decrease the stability of primary productivity and alter the strength of underlying biotic mechanisms (Muraina et al., 2021).

Top-down effects on community stability are also likely mediated by biotic stability mechanisms, but few studies have delineated these mechanistic relationships (Qin et al., 2019; Fu et al., 2020; Liang et al., 2021; Campana et al., 2022). It is well-known, however, that top-down pressure can directly alter both community stability and structure (Sankaran et al., 2004; Beck et al., 2015; Riginos et al., 2018; Ganjurjav et al., 2019). Changes in top-down effects, such as herbivore loss or a reduction of historic herbivory levels, can cause slow, directional compositional change (Porensky et al., 2016, 2017; Riginos et al., 2018), potentially shifting the importance of different biotic mechanisms. For instance, herbivores may reduce asynchronous dynamics by opening space that allows other species to increase synchronously (Porensky et al., 2013), or by limiting temporary pulses of dominant species (Mortensen et al., 2018) that could have produced asynchronous dynamics. Additionally, herbivores may preferentially consume abundant species, thereby freeing resources and space for new species (Hillebrand et al., 2007; Porensky et al., 2013; Borer et al., 2014; Koerner et al., 2018; Mortensen et al., 2018); these effects would reduce dominant species population stability while simultaneously strengthening diversity-associated portfolio effects. Herbivore abundance and the functional composition of herbivore communities also likely influence biotic stability mechanisms, as variation in herbivore pressure and identity each determine the extent of herbivore impacts on plant community composition (Augustine and McNaughton, 1998; Fynn and O'Connor, 2000; Veblen et al., 2016; Riginos et al., 2018).

Most plant communities are simultaneously subject to both bottom-up and top-down control, and we must consider how ecological stability and biotic stability mechanisms respond to such drivers in tandem. While drought and herbivory may both limit plant production and enhance the variability of production through time (Gill, 2007; Craine et al., 2012; Aroca, 2013; Grant et al., 2014; Muthoni et al., 2014), they do so in different ways, and thus their effects on biotic stability mechanisms may sometimes be aligned and sometimes different. Drought occurs episodically in semi-arid grasslands and savannas and affects all species simultaneously, while herbivory continually operates, but may have greater effects on certain plant species than others due to selective herbivory and the diets of different herbivore guilds (Augustine and McNaughton, 1998). We expect both types of disturbance to increase synchrony: fluctuations in year- to year rainfall can drive synchronous dynamics (Tredennick et al., 2017), while herbivory may enhance synchrony by freeing space or resources for less abundant species to increase simultaneously (Porensky et al., 2013). We also expect that drought and herbivory will both negatively affect the population stability of dominant species, as competitive dominant species can be especially drought-sensitive (Kardol et al., 2010), and as herbivores preferentially consume all of the dominant species in this system (Odadi et al., 2013). While we expect drought and herbivory will produce similar effects on synchrony and population stability, we expect that their effects on species diversity and the portfolio effect will be opposing. Specifically, selective herbivory of the dominant species may provide opportunities for subordinate species to thrive and thus increase richness (Porensky et al., 2013; Koerner et al., 2018; Bråthen et al., 2021). On the other hand, we expect the reduction in water availability with drought will reduce species richness (Adler and Levine, 2007; Harrison et al., 2018). As a result, the combination of drivers may jointly sum to produce a greater effect on stability than either alone when they have similar effects on a biotic mechanism, or they may buffer impacts of the other if their effects on a particular mechanism differ.

Here, to investigate how drought and herbivory affect stability and its underlying mechanisms, we leverage the Kenya Long-Term Exclosure Experiment (KLEE), which has measured plant responses to six levels of herbivory from experimental combinations of domestic and wild large mammalian herbivores across 22 years, including multiple drought events. Because drought is a temporal phenomenon, we calculated stability and biotic mechanisms over a moving window and assessed stability at different temporal scales. We first hypothesized that (H1) drought and herbivory would both reduce the stability of herbaceous cover (Figure 1). Second, we hypothesized that (H2) the importance of different biotic stability mechanisms would depend on drought and herbivore pressure (Figure 1). Specifically, we expected that, (2a) Asynchronous dynamics would decrease with drought and herbivore pressure, (2b) drought and herbivory would reduce the population stability of the palatable, dominant species, and (2c) species richness and the portfolio effect would decrease with drought but increase with the intensity of herbivory. Finally, we hypothesized that (H3) the effects of drought and herbivory on stability would be mediated by these biotic mechanisms (Figure 1). Consistent with the broader literature on biotic stability mechanisms, we hypothesized that (3a) stability would increase with increasing asynchronous dynamics (Valencia et al., 2020), (3b) stability would increase with dominant species population stability (Sasaki and Lauenroth, 2011; Yang et al., 2017), and (3c) stability would increase with species richness and the portfolio effect (Thibaut and Connolly, 2013).

Total cover fluctuated through time after an initial recovery from pre-treatment levels (Figure 2B). Treatment differences in total cover remained mostly consistent through time, and treatments followed the same fluctuations, with sharp drops in total cover during droughts, even in plots with few (W) or no (O) large herbivores (non-excluded herbivores included rodents, hares, and small <15kg ungulates). Species richness also varied through time (Figure 2C). It was not consistently ordered by herbivore pressure, although when richness sharply increased after dry periods, treatments that included wildlife had the greatest plant species richness (Figure 2C).

Unraveling the determinants of ecological stability is a core theoretical and empirical pursuit in ecology with important implications for land management and conservation in a changing world. Here, we explored the effects of herbivory and episodic drought on stability directly and as mediated through biotic stability mechanisms, using a 22-year time series of herbaceous cover data from an herbivore exclusion experiment in a semi-arid savanna. We analyzed stability and biotic mechanisms over a moving window, allowing us to incorporate the effects of periodic drought into temporal stability analyses and understand the top-down and bottom-up drivers of stability through time. Drought and herbivory both directly decreased the temporal stability of herbaceous cover (Figure 3) but varied in their effects on mediating biotic mechanisms. Both drought and herbivory increased synchrony (a destabilizing dynamic) while herbivory, but not drought, reduced the population stability of dominant species (Figure 4). Although both drivers affected species richness, and in contrasting ways (Figures 4E,F), richness was not significantly associated with overall stability (Figure 5). Overall, the direct effects of drought and herbivory on stability were smaller in magnitude than their indirect effects mediated through biotic stability mechanisms (Figure 6).

Grasslands and savannas are disturbance-dependent systems that are frequently described by non-equilibrium dynamics (Higgins et al., 2000; Fuhlendorf and Engle, 2001; Bond, 2008; Riginos et al., 2018) and our findings highlight the dynamic nature of this savanna rangeland system. Here we found that the direct effects of both bottom-up and top-down drivers structured stability. Specifically, drought and herbivory reduced the temporal stability of total herbaceous cover in an additive fashion, supporting H1 (Figure 3). The relationship between herbivory and stability was particularly strong, and treatments appeared to pair into groups of O and W, MW and C, WC and MWC (Figure 3B). Overall, the reductions in stability were largely driven by a reduction in the mean cover (Supplementary Figures S5A, S6A). Although both drivers also reduced the variance in absolute terms (Supplementary Figures S5B, S6B), the relative variance increased because we defined stability as the mean total cover divided by the standard deviation. Drought reduces the total amount of biomass as water stress limits production (Craine et al., 2012; Aroca, 2013; Grant et al., 2014), and also introduces greater variability to the system when production tracks rainfall fluctuations through time (La Pierre et al., 2011). At the same time, temporal variation in herbivore behavior and population dynamics may either buffer or enhance variability in vegetative cover (Augustine and McNaughton, 1998; Fynn and O'Connor, 2000; Abraham et al., 2019; Staver et al., 2019). Herbivores limit mean total cover by removing plant tissue (Gill, 2007; Muthoni et al., 2014), and may either suppress production by reducing photosynthetic potential and removing nutrients (Sitters et al., 2020) or stimulate productivity when plants compensate for herbivore damage by increasing growth rates (Strauss and Agrawal, 1999; Agrawal, 2000; Knapp et al., 2012). While there is evidence that herbivores increase productivity in this system (Charles et al., 2017), the effect of this on stability was outweighed by tissue removal. Herbivore identity and stocking rate also mattered: cattle and mesoherbivores were the only groups that significantly reduced stability, and cattle had a stronger effect likely because they were stocked at higher biomass densities (4,740 vs. 730 kg/km²). This is consistent with previous findings that total herbivore pressure is a stronger predictor of productivity and shifts in community composition than herbivore identity, but individual herbivore types are more likely to drive variation in productivity across space and time and changes in individual species (Veblen et al., 2016; Charles et al., 2017; Wells et al., 2021).

Although asynchronous dynamics are commonly observed in terrestrial systems (Zhao et al., 2020; but see Houlahan et al., 2007) and are often a key stabilizing mechanism (Valencia et al., 2020), we observed instead varying levels of synchronous dynamics (defined as variance ratio >1) which appeared to drive the dynamic nature of vegetative cover in this system (Figures 4A,B). In other words, under no conditions did species asynchrony stabilize herbaceous cover per se, but there was variation in the degree to which synchrony was a destabilizing dynamic, associated with drought and herbivory. Both drought and herbivory indirectly reduced stability by increasing synchrony (Figures 4A,B): synchronous dynamics increased during time periods with more frequent or severe droughts and in communities experiencing more herbivory, supporting H2a. Further, these indirect effects were similar to or greater in magnitude than the direct effect of either driver (Figure 6, Table 5). The finding with respect to drought does contrast with other findings across precipitation gradients (a spatial comparison rather than a temporal one), which find that asynchronous (compensatory) dynamics are strongest in places with high precipitation variability–implying that precipitation fluctuations can decrease synchrony rather than increasing it as we found here (Hallett et al., 2014). Dominant species were on average less synchronous than the full community (Figure 4B), indicating that subordinate species also substantially contribute to community synchrony. This effect is likely attributed to a reduction in palatable dominant species under higher herbivory, which creates space for subordinate species to fluctuate synchronously in response to rainfall variability (Porensky et al., 2013).

Counter to the effects of synchrony, dominant species population stability promoted community stability, especially at low levels of herbivory (Figures 5B, 6). Herbivores commonly influence community structure by preferentially consuming dominant species (Hillebrand et al., 2007; Porensky et al., 2013; Borer et al., 2014; Koerner et al., 2018; Mortensen et al., 2018). We found that the population stability of dominant species declined with increased herbivory, particularly when cattle or mesoherbivores were present. In this system, herbivores consume all of the dominant species, particularly B. lachnantha and T. triandra (Odadi et al., 2013), likely contributing to the reduction in average population stability. In particular, B. lachnantha showed large increases when released from herbivore pressure (Veblen et al., 2016). Contrary to expectations, the population stability of dominant species as a whole did not respond to drought. However, total herbaceous cover and stability did decline during drought periods even as the stability of dominant species did not, implying that drought-related reductions in stability result more from the response of the community as a whole than from the sensitivity of dominant species. Overall, the stability of total herbaceous cover correlated well with the population stability of dominant species, consistent with previous findings (Grman et al., 2010; Liang et al., 2021).

Although the portfolio effect is often noted as important in determining stability, where more diverse communities have more stable ecosystem properties overall (Yachi and Loreau, 1999; Hautier et al., 2015; Isbell et al., 2015; Loreau et al., 2021), we did not find evidence for this here (Figures 4E,F), contrary to H3c that species richness would be positively correlated with temporal stability. Indeed, an increasing number of studies find that asynchrony and dominant species stability have a greater influence on stability than richness (Hallett et al., 2014; Valencia et al., 2020; Muraina et al., 2021). The portfolio effect predicts that aggregate ecosystem properties will vary less in diverse communities due to some form of compensatory dynamics and thus, species richness should reduce synchrony (Valencia et al., 2020; Loreau et al., 2021). We included this relationship in our structural equation models but when this pathway was significant (only in the model including cattle herbivory) richness increased synchrony and thus had a small negative indirect effect on stability (Figure 6C). This phenomenon is likely explained by richness fluctuations in response to the combination of bottom-up and top-down drivers. Species richness declined with drought, a signal that was weakly detected over a 10-year moving window (Figure 4E) but apparent in the time series (Figure 2C), and increased transiently when rain followed after drought (Figure 2C), especially where herbivory opened space for rarer species (Porensky et al., 2013). These transient increases in richness (of rarer species) may contribute to some of the observed synchrony patterns, especially as dominant species were on average less synchronous than the full community (Figure 4B).

Drivers of stability fluctuate through time, which may produce timescale dependent effects on stability (Zelnik et al., 2018; Clark et al., 2021; Ross et al., 2021). To address the potential effects of timescale on our analyses, we considered stability and its underlying mechanisms at three timescales of 5-, 10-, and 15-year moving windows. In particular, synchronous dynamics are timescale dependent phenomena and our results show that synchrony dominated at shorter timescales (Supplementary Figures S2B, S7), consistent with evidence in grasslands that synchrony is stronger at short timescales, while asynchronous dynamics often dominate at longer timescales (Zhao et al., 2020; Shoemaker et al., 2022). Despite this, stability declined when calculated over longer timescales, likely reflecting the intuitive outcome that including more years in the calculation leads to a higher chance of including more variation. Even the effects of drought, an inherently temporal phenomenon, did not change substantially with timescale (Supplementary Figure S8). The relationship between stability and synchrony remained unchanged by timescale (Supplementary Figures S3A–C). We similarly found no difference in the relationships between population stability or richness and stability at different timescales (Supplementary Figures S3D–I), showing that although temporal variation in precipitation amounts as well as herbivore abundance, types, or intensity could alter the importance of biotic stability mechanisms depending on the timescale of observation, they did not here. Therefore, while synchrony was timescale dependent and the timescale of calculation influenced the magnitude of stability values, timescale did alter the effects of herbivore treatments or drought events on stability and biotic mechanisms.

Delving into the role of both bottom-up and top-down processes in structuring stability is critical to understanding the full picture, but the majority of theoretical and empirical work on this topic has focused on variation across bottom-up gradients only (Grman et al., 2010; Hallett et al., 2014; Xu et al., 2015; Hautier et al., 2020). In examining an herbaceous community subject to strong bottom-up (drought) and top-down (herbivory) limitation, we found that these stressors affect biotic mechanisms of stability in sometimes similar and sometimes contrasting ways. Drought and herbivory both reduced the stability of total herbaceous cover by increasing species synchrony, suggesting a commonality of mechanism. In both cases, the more stressed a community was, whether by water limitation or herbivore consumption, the less capacity there was for species to differ asynchronously in their responses to variability. On the other hand, drought and herbivory differed in how they affected two other potential mediators of stability: herbivory reduced dominant species population stability while drought had little effect, and herbivory increased average species richness while drought decreased it. In an applied context, this means considering that the stability of plant biomass for livestock or other consumers is a function not only of how much production the abiotic environment can support, but also how grazing feeds back to influence stability both directly and indirectly.

Climate change is expected to lead to more extreme and variable precipitation patterns (Allan and Soden, 2008; Trenberth et al., 2014; Funk et al., 2021; IPCC, 2021), and at the same time, in savanna ecosystems such as the one studied here, large herbivores are being lost from some areas (Ripple et al., 2015) and restored (Seddon et al., 2014; Stalmans et al., 2019) or replaced by increased livestock in others (Du Toit and Cumming, 1999). Understanding how bottom-up and top-down drivers alter stability and stabilizing community dynamics will provide more ability to predict and adapt to these changing landscapes. In particular, in a future of increasing drought, we may expect to see an increase in synchronous dynamics in plant communities and a decrease in temporal stability as observed here, but these effects will in turn be mediated by changes in herbivore communities and their direct and indirect impacts on plant community dynamics. Livestock replacement of wildlife is generally at higher stocking densities than the wild herbivores lost, particularly in semi-arid parts of the African continent (Hempson et al., 2017). An increase in total herbivore pressure will likely amplify the effects of drought, although wild meso- and megaherbivores are likely to have smaller effects on stability than cattle. In considering the applications of ecological stability, it is important to note that there are many facets of stability. We measured the invariability of herbaceous cover here, which reflects aggregate patterns over time and can incorporate but not directly measure other facets of stability such as resistance to perturbations or the rate of recovery to initial conditions, i.e., resilience (Donohue et al., 2016). How “stability” itself aligns with management goals will depend on the values applied to a particular system; the constancy of total herbaceous cover, for example, may be a desirable quality for forage availability and erosion control (Fynn and O'Connor, 2000), while temporal variability and heterogeneity support other key ecosystem functions and dimensions of biodiversity (Fuhlendorf and Engle, 2001; Porensky et al., 2013; Bråthen et al., 2021). Overall, our ability to understand the sources of variation and stability in dynamic systems will be essential to predicting future patterns of stability and navigating a changing world.

Publicly available datasets were analyzed in this study. This data can be found here: https://doi.org/10.6084/m9.figshare.3175561.v1.

CE led analysis and writing. MC, CW, and LH assisted in analytical methods. TY, KV, DK, and LP designed, implemented, and maintained the KLEE plots. All authors contributed to the article, interpretation of results, manuscript envisioning, editing, and approved the submitted version.

The KLEE exclosure plots were built and maintained by grants from the James Smithson Fund of the Smithsonian Institution, the National Geographic Society (Grants 4691-91, 9106-12, and 9986-16) and the National Science Foundation (Grants LTREB DEB 97-26642, 03-16402, 08-16453, 12-56004, 12-56034, and 19-31224).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer KJ declared a shared affiliation with the author RL to the handling editor at the time of review.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.