Requirements for the Spatial Storage Effect Are Weakly Evident for Common Species in Natural Annual Plant Assemblages

Requirements for the spatial storage effect are weakly evident for common species in natural annual plant assemblages

1 ISAAC R. TOWERS , CATHERINE H. BOWLER,MARGARET M. MAYFIELD , AND JOHN M. DWYER School of Biological Sciences, The University of Queensland, Brisbane, Queensland 4072 Australia

Citation: Towers, I. R., C. H. Bowler, M. M. Mayfield, and J. M., Dwyer. 2020. Requirements for the spatial storage effect are weakly evident for common species in natural annual plant assemblages. Ecology 101 (12):e03185. 10.1002/ecy.3185

Abstract. Coexistence in spatially varying environments is theorized to be promoted by a variety of mechanisms including the spatial storage effect. The spatial storage effect promotes coexistence when (1) species have unique vital rate responses to their spatial environment and, when abundant, (2) experience stronger competition in the environmental patches where they perform better. In a naturally occurring southwest Western Australian annual plant system, we conducted a neighbor removal experiment involving eleven focal species growing in high-abundance populations. Specifically, we measured species’ fecundity across a variety of environmental gradients in both the presence and absence of neighbors. For the environmental variables that we measured, there was only limited evidence for species-specific responses to the environment, with a composite variable describing overstory cover and leaf litter cover being the best predictor of fecundity for a subset of focal species. In addition, although we found strong evidence for intraspecific competition, positive environment–competition covariance was only detected for one species. Thus, positive environment–competition covariance may not be as common as expected in populations of species growing at high abundance, at least when tested in natural assemblages. Our findings highlight the inherent limitations of using natural assemblages to study spatial coexistence mechanisms, and we urge empirical ecologists to take these limitations into account when designing future experiments. Key words: annual plants; environmental heterogeneity; modern coexistence theory; neighbor removal experiment; southwest Western Australia; spatial storage effect; York gum-Jam woodlands.

that long-term species coexistence depends on “stabiliz- INTRODUCTION ing” mechanisms that cause niche differences to out- It is well recognized that diversity tends to increase weigh average fitness differences. Stabilizing mechanisms rapidly as the spatial extent of observation expands effectively drive negative frequency dependent popula- (MacArthur and Wilson 1967). One of the main tion growth, such that species experience high per capita hypotheses invoked to explain this phenomenon is that a population growth when they are rare (i.e., at low fre- greater range of niche opportunities become available quency, or in their “invader” state), but limit themselves for species to occupy as a wider variety of environmental when they become abundant (i.e., at high frequency, or conditions are captured at broader scales (Hart et al. in their “resident” state). 2017). Indeed, with few exceptions, empirical studies There are three recognized spatial coexistence mecha- reveal a positive relationship between within-community nisms that operate at scales large enough to capture environmental heterogeneity and community-level spe- meaningful environmental heterogeneity, the spatial cies richness, suggesting that increasing environmental storage effect, spatial relative nonlinearity and growth- variation is a reasonable explanation for the scale depen- density covariance, and all require that species have dif- dence of diversity (Stein et al. 2014). However, the mech- ferential responses to the spatially varying environment anisms behind this relationship, although well developed (Chesson 2000a). In this study, we focus on the spatial in theory, remain to be widely tested in natural commu- storage effect (Chesson 2000a), which further requires nities. that each species, when in their resident state, experi- Chesson’s (2000b) modern coexistence theory pro- ences strong competition in patches that are environ- poses a series of diversity maintenance mechanisms mentally favorable. This environment–competition including some that operate in the presence of environ- covariance (covEC) is theorized to be stronger (more mental heterogeneity. The central tenet of the theory is positive) when species are abundant because of the increased resource uptake of conspecific individuals in Manuscript received 20 March 2020; revised 29 June 2020; preferred patches. The final requirement of the spatial accepted 20 July 2020. Corresponding Editor: Gordon A. Fox. storage effect is that the invader-state population growth 1 E-mail: [email protected]

Article e03185; page 1 Article e03185; page 2 ISAAC R. TOWERS ET AL. Ecology, Vol. 101, No. 12 is buffered from the doubly negative effect of being in a species being studied and the ability to detect if the poor patch (which is likely favorable for another species) mechanism is operating (Levins 1966). and the intense interspecific competition that results. In this study, we examine environmental and competi- However, unlike the temporal storage effect, which tive responses of eleven winter annual plant species in requires a long-lived or resilient life history to “buffer” woodland understories in southwest Western Australia. the negative effects of a poor year, buffering occurs in We relax the requirement for all species’ responses to be the spatial context automatically when contributions to measured in the same patches and instead locate patches population growth rate in good patches compensate for of each species at many positions along measurable envi- bad patches, assuming that individuals have the oppor- ronmental gradients that are strongly associated with tunity to disperse between patches (Chesson 2000a, species turnover. We also focus only on common species 2008). assumed to be near their resident state, as these are most Mathematically, the strength of the spatial storage likely to exhibit positive covEC indicative of the spatial effect is a product of (1) the difference in covEC between storage effect (Chesson et al. 2013). We address the fol- resident and invader states and (2) the magnitude of spa- lowing questions: (1) Are species environmental tially buffered population growth (Chesson et al. 2013). responses related to measured environmental variables, Coexistence is promoted when resident covEC is greater and if so, do relationships differ among species? (2) How in magnitude than invader-state covEC. Thus, it is not evident is positive covEC in abundant species assumed the sign (positive or negative) of the resident-state to be near their resident state? covEC that matters but its relative magnitude in relation to invader-state covEC. The ideal experiment to test the METHODOLOGY requirements of the spatial storage effect, then, would be to sow focal species across a heterogeneous environment Study system and selectively manipulate competitive neighborhoods such that each focal species experiences competition in The York-gum–jam woodlands were once common in different environments at both its resident and invader southwest Western Australia but now occupy only 3% of state (with other focal species at their resident state; Li their historic range, persisting as discrete remnant 2016). Although such an experiment is feasible for a patches of varying size and quality within a matrix of small number of species, obtaining enough data to suffi- pasture, wheat, and canola fields (Hobbs and Yates ciently quantify the mathematical components of the 2000). Occurring on infertile, sandy loam soils, York- spatial storage effect (and other spatial coexistence gum woodlands have a sparse tree overstory mechanisms) quickly becomes intractable as the number (Appendix S1: Fig. S1b) characterized by two species: of species increases. This logistical complexity may York gum (Eucalyptus loxophleba) and Jam (Acacia explain the dearth of empirical studies attempting to rig- acuminata). The woodland understorey is dominated by orously quantify spatial storage effects in diverse natural a high diversity and abundance of annual forbs and win- systems to date. But it is perhaps surprising how few ter-growing geophytes interspersed among scattered empirical studies have used simpler, albeit less rigorous shrubs, perennial tussock grasses (e.g., Austrostipa spp.) approaches, to detect signals of the spatial storage effect, and bare ground (Prober and Wiehl 2012). Herbaceous especially given its potential importance for maintaining plant alpha diversity can be as high as 29 species per diversity in spatially heterogeneous natural systems 0.09 m2 (J. M. Dwyer and M. M Mayfield, unpublished globally. data). The sparse distribution of the overstory creates Neighbor-removal experiments can test for the local-scale heterogeneity in shade, tree litter and fallen operation of spatial storage effects in natural assem- woody debris (among other environmental variables). blages (Sears and Chesson 2007). However, natural These factors are strongly associated with turnover in assemblages impose their own limitations in this con- plant diversity in the York-gum–jam woodlands (Wain- text. For example, it is often difficult or impossible to wright et al. 2017), making it a strong candidate system locate environmental patches (defined at a scale that to investigate the role of spatial environmental variation can be considered spatially homogenous) in which on species coexistence. more than a few focal species consistently co-occur, especially if species occupy segregated spatial niches. Experimental design In addition, establishing sufficient within-patch replication for many species in these patches is likely to To assess species’ responses to the environment with be a challenge. Thus, further methodological adjust- and without neighbors, we conducted a spatially nested ments are needed if the goal is to examine signatures neighbor removal experiment in West Perenjori Nature of spatial storage effects for a satisfying number of Reserve, Western Australia, Australia (29°28040″ S, species in realistically diverse natural assemblages, but 116°12000″ E) in the warmer, northern extent of the it is unclear how such compromises affect our ability York-gum–jam woodlands (Appendix S1: Fig. S1a). Our to detect if the mechanism is operating. In other decision to increase community representation beyond words, a trade-off likely exists between the number of previous field studies of the spatial storage effect (Sears December 2020 LIMITED EVIDENCE FOR THE STORAGE EFFECT Article e03185; page 3 and Chesson 2007) by examining eleven species made it TABLE 1. Rate of survival to reproductive maturity for the 11 impossible to assess the responses of all species to the focal species. very same patches (because all eleven species rarely co- Species Survival rate occurred at the patch scale in our natural assemblages). We therefore selected only species that were common Arctotheca calendula 0.97 and abundant and assessed their fecundity responses Daucus glochidiatus 0.99 individually in 16 patches per species, ensuring that the Gilberta tenuifolia 1 measured environmental space covered by the patches Hyalosperma glutinosum 0.98 was similar across species. While this design only exam- Medicago minima 1 ined one focal species per patch, it allowed us to include Monoculus monstrosus 0.87 Pentameris airoides 0.99 within-patch replication of thinned and unthinned Plantago debilis 0.99 plants, which strengthened some aspects of our statisti- Podolepis canescens 0.99 cal analysis (Sears and Chesson 2007). Trachymene cyanopetala 1 At the beginning of the 2018 growing season (late Velleia rosea 1 July), 16 patches (1 m2) for each focal species were located across an ~12-ha section of the York-gum woodlands occurring in the south of the reserve (Appendix S1: Fig. S1a). Patch size was selected to mini- facing north. Because of the sun’s daily path across the mize abiotic heterogeneity (Sears and Chesson 2007), northern sky during the growing season in our study sys- while accommodating multiple subplots. To ensure that tem, we processed only the northern half of canopy sufficient variation in the physical environment was cap- cover photographs in ImageJ (Abramoff et al. 2004) to tured, patches were systematically established across nat- estimate overstory cover percentage (%). Percent cover ural gradients of tree cover, sclerophyllous litter cover (%) of sclerophyllous leaf litter was estimated by overlay- and coarse woody debris (branches or logs > 5 cm diam- ing a 100-point grid on digital photos of each patch and eter). Within each patch, six subplots (7.5 cm radius, at counting “hits.” Coarse woody debris (CWD) in or least 15 cm apart) were established, centered on individ- immediately adjacent to the patch was recorded as a bin- uals of the focal species. Each subplot was then ran- ary variable, where branches or logs greater than 5 cm domly assigned to be a thinned or unthinned subplot diameter that were in contact with the ground were con- (three of each). sidered a “presence.” A soil sample (0–70 mm depth, In thinned subplots, the aboveground biomass of all excluding litter) was collected from each patch at the germinants within 7.5 cm of target individuals was care- beginning of the growing season to best represent avail- fully removed by hand. This neighborhood radius was able nutrients before plant uptake and leaching considered to be sufficiently large to minimize local occurred. Soil samples were air-dried and analysed plant–plant interactions based on a prior study in this (School of Agriculture and Food Sciences, The Univer- system (T. Martyn, unpublished data). Thinning occurred sity of Queensland) for extractable phosphorous (mg/ early in the season as soon as focal species could be reli- kg), extractable potassium (mg/kg), and plant-available ably identified. Unthinned subplots were not manipu- nitrogen (%). lated. At peak biomass, the identity and abundance of all neighbours within each unthinned subplot was Statistical analyses recorded. We monitored the survival to reproductive maturity Are species environmental responses related to measured and seed production of each focal individual. To assess environmental variables, and if so, do relationships differ seed production, thin mesh bags were fastened over among species?.—To first assess species-specific immature fruiting bodies to capture seeds. This occurred responses to the environment in the absence of neigh- shortly prior to seed release to minimize disturbance to bors, fecundity in thinned subplots for each species was the plant. For the majority of species, a target individ- modeled as a function of environmental conditions using ual’s responses to the environment was then recorded as generalized linear mixed-effects models (GLMMs) with its total seed production, excluding seeds that were a negative binomial error distribution and log link func- unfilled. In the case of P. airoides, the number of florets tion. Environmental variables included overstory cover, was recorded and then multiplied by two as a measure litter cover, coarse woody debris (CWD), extractable of fecundity as florets contain two seeds on average. Sur- phosphorous (P), extractable potassium (K), and plant- vival rates were high for all species (Table 1) so, instead available nitrogen (N). Due to collinearity between cer- of modeling survival as a separate vital rate, we instead tain environmental variables, we first conducted a global treated all individuals that died before reaching repro- principal component analysis (PCA) on all six variables. ductive maturity as having zero seed production. The global PCA revealed that soil variables were mostly Patch-level overstory cover was assessed by taking a loaded on a different axis than overstory cover and litter wide-angle digital photograph of tree canopies from the cover and so we ran a separate PCA for the soil variables center of each patch at 25 cm from the ground and and another for overstory cover and litter cover Article e03185; page 4 ISAAC R. TOWERS ET AL. Ecology, Vol. 101, No. 12

(Appendix 1: Fig. S2). Preliminary data exploration log link function. We excluded occasional occurrences of revealed that CWD was uncorrelated with the other vari- the parasitic vine Cuscata campestris when calculating ables and so we excluded CWD from the PCA to main- absolute neighbor abundance. Total neighbor abundance tain its binary distribution. We then extracted the first and the abundance of conspecifics was ln(fecundity + 1) principal component from each PCA to describe “soil transformed to improve model fit. fertility” and “overstory and litter cover,” respectively. Soil fertility, overstory and litter cover and CWD were How evident is positive covEC in abundant species included in each species’ fecundity model and patch was assumed to be near their resident state?.—We first tested included as a random effect. Quadratic relationships for positive covEC, where E is estimated as the average were also explored for each species and environmental ln-transformed fecundity in thinned subplots in a given variable. To reduce bias in parameter estimates for linear patch. C in the same patch is the average difference in terms, non-significant quadratic terms were removed in fecundity due to the presence of neighbors. However, a one-step model simplification process. We also mod- this approach uses the fecundity of thinned plants in the eled fecundity in unthinned subplots for each species in calculation of both E and C. The common sampling an intercept-only mixed-effects model (including plot as error this introduces must be statistically removed before a random effect) to compare species-average fecundity testing for covariance. As per Sears and Chesson (2007), (Table 2). we used maximum-likelihood to estimate values for the E and C using the following models: Estimating effects of neighbors on fecundity.—Overall ð Þ¼ þ e effects of neighbors on each focal species were ln yt;x;i Ex t;x;i assessed using mixed-effects GLMMs (negative binomial error distribution and log link). Treatment lnðyu;x;iÞ¼Ex Cx þ eu;x;i (thinned or unthinned) was the sole fixed effect and patch was included as a random effect in each model. Cx;j ¼ a þ bEx Positive responses to thinning when averaged across patches indicate that neighbors had a competitive where y is the fecundity of individuals, i, in thinned, t, effect on focal individuals while negative responses and unthinned, u, subplots nested within patch, x is the indicate that neighbors had a facilitative effect on observation-level error, and a and b are, respectively, the focal individuals. intercept and regression coefficient of the relationship We also used neighborhood composition data to investi- between C and E. Parameter estimates for a and b were gate density-dependent effects of local-scale competition. then calculated using orthogonal regression (Carroll This was done in part to check that our chosen focal spe- et al. 2006). In contrast to ordinary least squares, cies were abundant enough to limit themselves. Specifi- orthogonal regression assumes that error variance cally, we modeled fecundity in unthinned subplots as a occurs in both variables, rather than just in the depen- function of the total abundance of all neighbors as well as dent variable. It therefore requires an estimate of the the absolute abundance of conspecific individuals using within-patch variance for thinned and unthinned treat- GLMMs with a negative binomial error distribution and ments, which are expressed as the ratio h

TABLE 2. Parameter estimates from intercept-only and species-specific environmental GLMMs of fecundity in unthinned subplots.

Species k Tree overstory and litter cover Soil fertility Presence of CWD Arctotheca calendula 79.44 (54.23–116.36) 0.63** 0.29 1.19* Daucus glochidiatus 19.31 (17.05–21.86) 0.08 (0.32**) 0.04 0.42 Gilberta tenuifolia 81.66 (69.84–95.49) 0.06 0.01 0.14 Hyalosperma glutinosum 60.30 (50.26–72.34) 0.19 0.08 0.04 Medicago minima 11.94 (10.62–13.42) 0.09 0.18 0.18 Monoculus monstrosus 8.61 (7.03–10.55) 0.02 0.23 0.46 Pentameris airoides 234.15 (197.36–277.80) 0.18 0.05 0.43 Plantago debilis 46.89 (38.93–56.46) 0.07 0.27 0.61 Podolepis canescens 489.56 (416.90–574.89) 0.15 0.05 0.12 Trachymene cyanopetala 41.40 (35.69 –48.01) 0.16 0.23* 0.27 Velleia rosea 18.52 (16.12–21.27) 0.26* (0.20*) 0.15 0.16 Notes: The parameter: k is the estimated intercept value in the intercept-only mixed effects model. Values in brackets after intercept values represent one standard deviation around the parameter estimate of the intercept. Values under tree overstory and litter cover, soil fertility, and the presence of coarse woody debris (CWD) represent the fixed-effect parameter estimates, while values in brackets after the parameter estimates of linear terms represent the estimated quadratic term where it was significant. Fixed effects parameter estimates are log-transformed. *P ≤ 0.05, **P ≤ 0.01. December 2020 LIMITED EVIDENCE FOR THE STORAGE EFFECT Article e03185; page 5

n n Px Pi 2 intervals around the Deming slope estimate were y ; ; y u x i u;x inspected for each species to assess the significance of h ¼ x¼1 i¼1 s Pnx Pni slope differences. The biological interpretation of Dem- 2 yt;x;i yt;x ing regressions was consistent with the results from the x¼1 i¼1 maximum likelihood method described above and so we Error variance ratios ranged from 0.29 to 3.25 across refer herein to the results for maximum likelihood only the eleven focal species. Covariance between C and E (but see Appendix S1: Fig. S3 and Table S1). was then estimated as b multiplied by the variance in E. Statistical analysis was performed in R (R Core Team An F test was performed to determine whether b was sig- 2019), Deming regression was performed using the mcr nificantly different to 0 (a = 0.05) for each species. A package (Manuilova 2014) and mixed-effects modeling positive value for covEC is evidence that neighbors limit was performed using the lme4 package (Bates et al. 2015). fecundity more in favorable patches (Fig. 1), but impor- tantly, covEC need not be positive for the spatial storage RESULTS effect to be operating. The removal of common sampling error can substantially alter the slope of the E–C rela- Are species environmental responses related to measured tionship, such that the adjusted relationship may not fol- environmental variables, and if so, do relationships differ low the “raw” data when plotted. To facilitate among species? interpretation of results and how they relate to raw data, ’ we modeled relationships between the average ln-trans- Species fecundity in the absence of competition did not formed fecundity of thinned and unthinned plants in exhibit strong differential responses to measured environ- each patch using Deming regressions. Positive covEC is mental gradients. Out of 33 combinations of species and apparent when the Deming regression slope is signifi- environmental gradients, we found only five to be signifi- a = cantly less than 1. Analytically derived 95% confidence cant ( 0.05; Table 2). The composite variable describing overstory and litter cover, where high values represent shaded patches with high leaf litter cover and low values represent bare, open patches, was the significant environmental predictor in the majority of these relationships. A. calendula had a strong positive relationship, D. glochidiatus had a U-shaped relationship, and V. rosea had a hump-shaped but mostly negative response (Fig. 2). T. cyanopetala’s fecundity was negatively related to soil fertility while A. calendula’s was positively related to CWD (Appendix S1: Fig. S4).

Estimating effects of neighbors on fecundity Four out of 11 species exhibited significant responses in their fecundity to neighborhood thinning when averaged across plots and in all cases these responses were positive (i.e., neighbor removal had a positive effect indicating competition; Table 3). Similarly, species’ fecundities responded negatively to increasing neighbor density (conspecific and total) in all cases where the relationship was significant (Table 3). Conspecific neighbor density was significantly and negatively related to fecundity for FIG. 1. Conceptual illustration of positive covariance six species, indicating that these species were abundant between the favorability of the environment (response to environment) and the strength of competition (response to competi- enough to limit themselves (Appendix 1: Fig. S5; tion) when using maximum likelihood (see Methods). Response Table 3). Interestingly, the total abundance of neighbors to environment is estimated as the ln(fecundity) in thinned plots explained fecundity for only four species (Appendix 1: while response to competition is estimated as the ln(response Fig. S6), and only A. calendula responded significantly ratio) of fecundity in thinned subplots divided by unthinned to both the number of conspecific neighbors and the subplots. The dashed line is the fitted model when environment–competition covariance (E-C cov) is significant and posi- total abundance of neighbors. tive while the solid line is the fitted model when neighbors have no effect on fecundity. Dark gray fields represent areas of statistical space where the fecundity in unthinned plots is greater How evident is positive covEC in abundant species than in thinned plots (i.e., facilitation) while unshaded fields assumed to be near their resident state? represents areas of statistical space where the fecundity in thinned plots is greater than in unthinned plots (i.e., competi- For the majority of species, there was little to no evi- tion). [Color figure can be viewed at wileyonlinelibrary.com] dence of positive covariance between patch favorability Article e03185; page 6 ISAAC R. TOWERS ET AL. Ecology, Vol. 101, No. 12

a b c A. calendula D. glochidiatus V. rosea 60 800 40 50

600 30 40

30 400 20

20 Fecundity (thinned) Fecundity Fecundity (thinned) Fecundity 200 (thinned) Fecundity 10 10

0 0 −2 −1 0 1 2 −1 0 1 2 −1 0 1 2 Tree overstory and litter cover Tree overstory and litter cover Tree overstory and litter cover

FIG. 2. The effect of tree overstory and litter cover on the fecundity of Arctotheca calendula, Daucus glochidiatus, and Velleia rosea in thinned subplots. High values of overstory and litter cover represent shaded patches with high leaf litter cover while low values represent bare, open patches. Each point is the observed fecundity in a subplot. Light gray fields represent the 95% confidence interval. Lines represent the fitted slope estimate for each relationship.

TABLE 3. The average effect of neighbors on fecundity and the 4 effect of the total number of neighboring individuals and Competition number of intraspecific neighbors on fecundity in thinned subplots. 3 Species Average Total Intraspecifc Arctotheca calendula n.s. –*** –* Daucus glochidiatus n.s. n.s. n.s. Gilberta tenuifolia –** –*** n.s. 2 Hyalosperma glutinosum n.s. n.s. –*** Medicago minima n.s. –** n.s. Monoculus monstrosus n.s. n.s. n.s. 1 Pentameris airoides –*** n.s. –* –*** –** Plantago debilis n.s. Response to competition Podolepis canescens n.s. n.s. –*** Trachymene cyanopetala n.s. n.s. –* 0 Velleia rosea –* –** n.s. Facilitation Notes: The sign of the relationship is indicated when significant; n.s. indicates the relationship was not significant. 4.0 4.5 5.0 5.5 6.0 6.5 *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001. Response to environment

FIG. 3. The relationship between response to competition and the effect of competition. CovEC was significant and response to environment for Pentameris airoides. Response and positive for one common species, P. airoides, and to competition is estimated as the ln(response ratio) of plot-level then, only when maximum likelihood was used to esti- average fecundity in thinned subplots divided by unthinned sub- mate slope parameters (covEC = 0.21, P = 0.04, Fig. 3). plots (y-axis) while the response to environment is estimated as the plot-level average fecundity in thinned subplots (x-axis). The Hyalosperma glutinosum, Plantago debilis, Podolepis line shows the maximum likelihood fit after correcting for com- canescens, and Trachymene cyanopetala showed moder- mon sampling error between the estimates of competition C and ately negative covEC, although in no cases was this sig- environment E. Dark gray fields represent areas of statistical nificant (Table 4). space where the fecundity in unthinned plots is greater than in thinned plots (i.e., facilitation) while unshaded fields represents areas of statistical space where the fecundity in thinned plots is greater than in unthinned plots (i.e., competition). DISCUSSION The spatial storage effect is theorized to be an important mechanism stabilizing coexistence in spatially vary- eleven focal species yielded a number of important find- ing environments yet empirical tests of the key ings. Firstly, although we observed moderate differentia- requirements of this mechanism are extremely rare. Our tion in fecundity responses to the measured environment experimental manipulation of plant neighborhoods for some species, fecundities of more than one-half of across multiple orthogonal environmental gradients for the focal species were not significantly related to any December 2020 LIMITED EVIDENCE FOR THE STORAGE EFFECT Article e03185; page 7

TABLE 4. The slope estimate, b, between the response to improve soil moisture retention (Walsh and Voigt 1977). – competition, C, and response to environment, E, E C Conversely, light interception by the overstory and a covariance (b 9 Var(E)), and the significance of b from maximum likelihood analysis. dense leaf litter layer (Scholes and Archer 1997) as well as allelopathic toxins leached from sclerophyllous litter Species bE–C covariance P (May and Ash 1990) may limit growth in the under- Arctotheca calendula 0.13 0.37 0.46 storey. These potential effects of overstory and litter Daucus glochidiatus 0.47 0.06 0.30 cover may explain the contrasting responses that we Gilberta tenuifolia 0.15 0.03 0.77 observed between A. calendula and V. rosea although Hyalosperma glutinosum 0.13 0.04 0.69 further work is needed to tease apart the individual Medicago minima 0.50 0.05 0.42 mechanisms. Monoculus monstrosus 0.35 0.17 0.17 It is important to acknowledge that weak evidence for Pentameris airoides 1.30 0.21 0.04 spatial niche differentiation may simply reflect our Plantago debilis 0.27 0.09 0.44 choice of environmental variables or the scale over which Podolepis canescens 0.14 0.02 0.87 we studied them. Indeed, the fact that we could not fre- Trachymene cyanopetala 0.09 0.03 0.67 quently locate all eleven of our species in the same Velleia rosea 0.14 0.02 0.74 patches suggests that species do exhibit unique responses to the environment, but the environmental variables we measured did not explain this interspecific variation in examined environmental variables. Secondly, we found responses. Nevertheless, the environmental variables we evidence for positive covEC for only one species, despite measured are strongly associated with community turn- strong evidence of intraspecific competition for most over in these York-gum woodlands (Prober and Wiehl species. These findings suggest that positive covEC, at 2012, Wainwright et al. 2017). Thus, we had a reasonable least when mediated exclusively by species’ fecundity a priori expectation that species would exhibit a greater responses in this natural system, does not commonly diversity of responses. As highlighted previously, it is occur for species near their resident state. also possible that weak species-specific responses were recorded as a result of locating focal individuals in naturally occurring assemblages. As a consequence, we may Are species environmental responses related to measured not have captured positions on environmental gradients environmental variables, and if so, do relationships differ where species have been “pre-filtered” from local patches among species? over multiple generations of poor performance In order for coexistence to be stabilized by the spatial (HilleRisLambers et al. 2012). Again, sowing experi- storage effect species must exhibit partially uncorrelated ments across complete gradients may capture species’ demographic responses to their environment (Sears and responses more completely in future studies. Finally, it is Chesson 2007). For four out of the 11 focal species inves- possible that weak species-specific responses were the tigated, we found significant relationships between result of the benign, above-average rainfall conditions fecundity and the environment, and the nature of these that occurred throughout 2018 when this experiment relationships differed by species. Consequently, we have was conducted. Environmental factors that can amelio- some evidence that the first requirement of the spatial rate drought stress, such as the presence of CWD and storage effect is satisfied for a subset of the species stud- the amount of overstory and litter cover, may play a ied in this system. The insignificant relationships much smaller role in mediating species’ performances in observed for the remaining species suggest that system wet years (Helman et al. 2017). Given that the spatial wide, spatial niche differentiation is either weak, involves storage effect is dependent on partially uncorrelated untested environment factors or only operates across lar- responses to patch environments, this raises an interest- ger spatial scales than were tested here (Hart et al. ing avenue for future research: does the strength of the 2017). Alternatively, and most likely, weak responses to spatial storage effect vary systematically with climate the environment may simply reflect the limitations of through time as a spatiotemporal dynamic? experiments in natural assemblages. Sparsely distributed woody plant species often play an How evident is positive covEC in abundant species important role in structuring understorey community assumed to be near their resident state? composition in semiarid ecosystems (Facelli and Brock 2000, Prober and Wiehl 2012, Helman et al. 2017, Tes- In the only other field study to quantify the second sema and Belay 2017). In the same way, when species in requirement of the spatial storage effect, Sears and the present study responded significantly to the environ- Chesson (2007) found that the reserve-scale yield of ment, they tended to respond to the combined effect of Erodium cicutarium (an abundant species assumed to be overstory and litter cover. Overstory cover can facilitate in its resident state) was significantly limited by environ- survival and performance in the understorey by limiting ment–competition covariance, suggesting that spatial direct solar irradiation (Valladares et al. 2016), conse- storage effect may promote coexistence in this system. quently alleviating water stress, while leaf litter cover can Consistent with Sears and Chesson (2007), we found Article e03185; page 8 ISAAC R. TOWERS ET AL. Ecology, Vol. 101, No. 12 evidence for positive covEC for one common species, (Appendix 1: Fig. S5), which could potentially explain the invasive annual grass P. airoides. Positive covEC is the weak covariance we observed for these species. expected to arise from an increase in per capita We focused exclusively on spatial variation in species’ intraspecific competition under preferable environmen- fecundity when testing the key requirements of the spa- tal conditions. Due to the nature of our experiment, tial storage effect. However, annual plant species’ long- however, we were unable to explore whether this was the term population growth rates are a function of multiple mechanistic basis for the positive covEC that we demographic rates and covariation in any of these rates observed for P. airoides. Our findings for P. airoides are with the environment could lead to stabilization via the thus consistent with Sears and Chesson (2007) by show- spatial storage effect, so long as it causes species to expe- ing positive covEC for at least one abundant species. By rience greater competition in the patches where they are studying more species, however, we were able to show most successful (Chesson 2000a). For example, species- that this phenomenon was not nearly as common as we specific variation in germination rates has been shown expected. As highlighted previously, manipulation of to be crucial to the operation of the temporal storage conspecific competitor density to predetermined densi- effect in winter annuals in the Sonoran desert (Angert ties through seed addition and selective thinning, paired et al. 2009) and we predict that spatial variation in ger- with completely thinned neighborhoods, could allow mination rates may be an important driver of spatial future studies to isolate variation in per-capita coexistence mechanisms in many systems including the intraspecific competition across environmental gradients York-gum woodlands. Survival in the seed bank is (Hart et al. 2017). another potentially important, but often overlooked, For the remaining ten focal species we observed weak demographic rate and could vary with patch type for a covEC despite evidence that intraspecific competition number of reasons. For example, seed pathogen load is reduced fecundity in most (six) of these species. Thus, known to vary with microsite conditions such as litter most species experienced negative frequency depen- cover in a plant density-independent manner (Beckstead dence, but its effects were independent of the measured et al. 2012) and this could disproportionately reduce spatial environment. This suggests that stabilizing mech- seed survival for species with poor pathogen resistance. anisms operating at the scale of plant neighborhoods Species-specific seed predation post-dispersal may also may also be important in promoting coexistence (Adler be spatially dependent (von Euler et al. 2014). For exam- et al. 2018). ple, many studies have shown that rodents, which are The strength of storage effects are known to vary also seed predators in our system, tend to forage in shel- among systems. When examining the temporal storage tered microsites. Finally, although we observed very high effect, Adler et al. (2009) found weak evidence for envi- rates of seedling survival in this study, variation in sur- ronment–competition covariance for all four species vival to reproductive maturity, especially at very early tested in North American sagebrush steppe despite find- life stages, may play an important role in promoting ing strong evidence for species-specific responses to the coexistence in a manner independent of fecundity. environment. In contrast, Adler et al. (2006) found Annual plant models that allow all vital rates to explic- strong evidence for environment–competition covariance itly vary with the environment may provide a more accu- in three North American prairie grass species. Our find- rate and mechanistic perspective on the effect of spatial ings add key novel information about the commonality variation on coexistence (Bimler et al. 2018). of the requirements of the spatial storage effects in a It is also important to acknowledge that other spatial well-studied annual plant system of the sort in which this coexistence mechanisms may be operating in this system. mechanism is widely expected to operate. We hope that In addition to the spatial storage effect, relative non-lin- the results from this study encourage further tests of the earity of competition and growth–density covariance key requirements of the spatial storage effect to deter- can promote coexistence as a consequence of spatial mine whether similarly low incidences of positive covEC variation (Chesson 2000a, Snyder and Chesson 2003, are generalizable to other systems. In particular, we urge Snyder and Chesson 2004, Holt and Chesson 2016). For future studies to explicitly test resident-invader differ- example, in the context of temporal variation, recent ences in covEC for multi-species communities where studies indicate that relative non-linearity of competition possible, although the spatial design of such an experi- plays a much greater role than the storage effect in pro- ment still represents a major empirical challenge. moting species coexistence in a variety of systems includ- Despite our aim to select only abundant focal species, ing different grasslands and nectar yeasts (Letten et al. the results of our neighborhood competition analyses 2018, Hallett et al. 2019, Zepeda and Martorell 2019). In (Appendix 1: Fig. S5) suggest that some were not locally a similar manner, spatial relative nonlinearity of compe- abundant enough to create the competitive conditions tition and growth–density covariance have not yet been required for negative frequency dependence. For exam- empirically studied but they may play an important role ple, M. monstrosus and D. glochidiatus had maximum in promoting coexistence in spatially varying environ- conspecific neighbor counts of three and five respec- ments (Snyder and Chesson 2003). Thus, we urge future tively, compared to conspecific neighbor counts of over empirical studies to consider these overlooked mecha- 40 for P. airoides, H. glutinosum, V. rosea, and P. debilis nisms. December 2020 LIMITED EVIDENCE FOR THE STORAGE EFFECT Article e03185; page 9

ACKNOWLEDGMENTS vegetation of open woodlands of south Australia. Ecography 23:479–491. Thanks go to the Perenjori Shire Council, Western Australia Hallett, L. M., L. G. Shoemaker, C. T. White, and K. N. Sud- for accommodating our research team while in the field. Thanks ing. 2019. Rainfall variability maintains grass-forb species also go to Kimberley Riddell, Eilish McMaster, Wenting Su, coexistence. Ecology Letters 22:1658–1667. Olivia Oltrogge, Maylon Bimler, Victoria Reynolds, Felipe Hart, S., J. Usinowicz, and J. Levine. 2017. The spatial scales of Suarez, Abigail Pastore, and the Dwyer and Mayfield Lab for species coexistence. Nature Ecology and Evolution 1:1066– assistance in the field and the laboratory and for helpful discus- 1073. sions. I. R. Towers and C. H. Bowler conducted the field experi- Helman, D., S. Leu, and A. Mor-Mussery. 2017. Contrasting ment and prepared the data, I. R. Towers, J. M. Dwyer, and M. effects of two Acacia species on understorey growth in a dry- M. Mayfield developed the concept of the study, I. R. Towers lands environment: Interplay of canopy shading and litter conducted the statistical analyses, and I. R. Towers, J. M. interference. Journal of Vegetation Science 28:1140–1150. Dwyer, M. M. Mayfield, and C. H. Bowler wrote the manu- HilleRisLambers, J., P. B. Adler, W. S. Harpole, J. M. Levine, script. and M. M. Mayfield. 2012. Rethinking community assembly through the lens of coexistence theory. Annual Review of Ecology, Evolution, and Systematics 43:227–248. LITERATURE CITED Hobbs, R. J., and C. J. Yates. 2000. Temperate eucalypt wood- Abramoff, M. D., P. J. Magalhaes,~ and S. J. Ram. 2004. Image lands in Australia : biology, conservation, management and processing with ImageJ. Biophotonics International 11:36–42. restoration. Surrey Beatty & Sons, Chipping Norton, New Adler, P. B., J. HilleRisLambers, P. C. Kyriakidis, Q. Guan, and South Wales, Australia. J. M. Levine. 2006. Climate variability has a stabilizing effect Holt, G., and P. Chesson. 2016. Scale-dependent community on the coexistence of prairie grasses. Proceedings of the theory for streams and other linear habitats. American Natu- National Academy of Sciences USA 103:12793–12798. ralist 188:E59–E73. Adler, P. B., J. HilleRisLambers, and J. M. Levine. 2009. Weak Letten, A. D., M. K. Dhami, P.-J. Ke, and T. Fukami. 2018. effect of climate variability on coexistence in a sagebrush Species coexistence through simultaneous fluctuation-depen- steppe community. Ecology 90:3303–3312. dent mechanisms. Proceedings of the National Academy of Adler, P. B., D. Smull, K. H. Beard, R. T. Choi, T. Furniss, A. Sciences USA 115:6745–6750. Kulmatiski, J. M. Meiners, A. T. Tredennick, and K. E. Levins, R. 1966. The strategy of model building in population Veblen. 2018. Competition and coexistence in plant commu- biology. American Scientist 54:421–431. nities: intraspecific competition is stronger than interspecific Li, Y. 2016. Biodiversity in a dynamic world: how environmen- competition. Ecology 21:1319–1329. tal variability influences coexistence between introduced and Angert, A. L., T. E. Huxman, P. Chesson, and D. L. Venable. native species. Dissertation. University of Arizona, Tuscon, 2009. Functional tradeoffs determine species coexistence via Arizona, USA. the storage effect. Proceedings of the National Academy of MacArthur, R. H., and E. O. Wilson. 1967. The theory of island Sciences USA 106:11641–11645. biogeography. Princeton University Press, Princeton, New Bates, D., M. Machler,€ B. Bolker, and S. Walker. 2015. Fitting Jersey, USA. linear mixed-effects models using lme4. Journal of Statistical Manuilova, E., and A. Schuetzenmeister. 2014. mcr: method Software 67. comparison regression. R package version 1.2.1. https://cran. Beckstead, J., L. E. Miller, and B. M. Connolly. 2012. Direct r-project.org/web/packages/mcr/mcr.pdf and indirect effects of plant litter on a seed–pathogen interac- May, F., and J. Ash. 1990. An assessment of the allelopathic poten- tion in Bromus tectorum seed banks. Seed Science Research tial of Eucalyptus. Australian Journal of Botany 38:245–254. 22:135–144. Prober, S. M., and G. Wiehl. 2012. Relationships among soil Bimler, M. D., D. B. Stouffer, H. R. Lai, and M. M. Mayfield. fertility, native plant diversity and exotic plant abundance 2018. Accurate predictions of coexistence in natural systems inform restoration of forb-rich eucalypt woodlands. Diversity require the inclusion of facilitative interactions and environ- and Distributions 18:795–807. mental dependency. Journal of Ecology 106:1839–1852. R Core Team. 2019. R: A language and environment for statisti- Carroll, R. J., D. Ruppert, L. A. Stefanski, and C. M. Craini- cal computing. R Foundation for Statistical Computing, ceanu. 2006. Measurement error in nonlinear models. CRC Vienna, Austria. www.R-project.org Press, Boca Raton, Florida, USA. Scholes, R. J., and S. R. Archer. 1997. Tree-grass interactions in Chesson, P. 2000a. General theory of competitive coexistence in savannas. Annual Review of Ecology and Systematics 28:517– spatially-varying environments. Theoretical Population Biol- 544. ogy 58:211–237. Sears, A. L. W., and P. Chesson. 2007. New methods for quanti- Chesson, P. 2000b. Mechanisms of maintenance of species diver- fying the spatial storage effect: an illustration with desert sity. Annual Review of Ecology and Systematics 31:343–366. annuals. Ecology 88:2240–2247. Chesson, P. 2008. Quantifying and testing species coexistence Snyder, R. E., and P. Chesson. 2003. Local dispersal can facili- mechanisms. Pages 119–164 in F. Valladares, A. Camacho, A. tate coexistence in the presence of permanent spatial hetero- Elosegui, C. Gracia, M. Estrada, J. C. Senar and J. M. Gili, geneity. Ecology Letters 6:301–309. editors. Unity in diversity: reflections on ecology after the Snyder, R. E., and P. Chesson. 2004. How the spatial scales of dis- legacy of Ramon Margalef, Fundacion BBVA, Bilbao. persal, competition, and environmental heterogeneity interact Chesson, P., N. J. Huntly, S. H. Roxburgh, M. Pantastico-Cal- to affect coexistence. American Naturalist 164:633–650. das, and J. M. Facelli. 2013. The storage effect: definition and Stein, A., K. Gerstner, and H. Kreft. 2014. Environmental tests in two plant communities. Pages 11–40 in C. K. Kelly, G. heterogeneity as a universal driver of species richness across A. Fox, and M. G. Bowler, editors. Temporal dynamics and taxa, biomes and spatial scales. Ecology Letters 17:866–880. ecological process. Cambridge University Press, Cambridge, Tessema, Z., and E. Belay. 2017. Effect of tree species on under- UK. story vegetation, herbaceous biomass and soil nutrients in a Facelli, J. M., and D. J. Brock. 2000. Patch dynamics in arid semi-arid savanna of Ethiopia. Journal of Arid Environments lands: localized effects of Acacia papyrocarpa on soils and 139:76–84. Article e03185; page 10 ISAAC R. TOWERS ET AL. Ecology, Vol. 101, No. 12

Valladares, F., L. Laanisto, U.€ Niinemets, and M. A. Zavala. gradients on annual plant community structure. Biological 2016. Shedding light on shade: ecological perspectives of Invasions 19:479–491. understorey plant life. Plant Ecology & Diversity 9:237–251. Walsh, R. P. D., and P. J. Voigt. 1977. Vegetation litter: An von Euler, T., J. Agren, and J. Ehrlen. 2014. Environmental con- underestimated variable in hydrology and geomorphology. text influences both the intensity of seed predation and plant Journal of Biogeography 4:253–274. demographic sensitivity to attack. Ecology 95:495–504. Zepeda, V., and C. Martorell. 2019. Fluctuation-independent Wainwright, C. E., J. M. Dwyer, and M. M. Mayfield. 2017. niche differentiation and relative non-linearity drive coexis- Effects of exotic annual grass litter and local environmental tence in a species-rich grassland. Ecology 100:e02726.

SUPPORTING INFORMATION Additional supporting information may be found in the online version of this article at http://onlinelibrary.wiley.com/doi/ 10.1002/ecy.3185/suppinfo