Beta diversity among prairie restorations increases with species pool size, but not through enhanced species sorting

Authors


Summary

  1. Understanding variation in community composition across space, or beta diversity, is of longstanding interest in ecology, yet the determinants of beta diversity remain poorly known. In part, this results from a lack of manipulative tests of hypothesized drivers. The size of species pools is one putative driver, but few studies have provided a direct test of this mechanism through manipulation of clearly defined species pools independent of local communities. Furthermore, we know little about underlying mechanisms, such as enhanced species sorting, or whether a species pool size-beta diversity relationship is scale-dependent or modified by environmental conditions.
  2. Here, we evaluate 29 prairie plant communities restored from bare soil with known species pools (seed mixes) to address those questions. To address the generality of beta diversity drivers across scales, we investigated how the size of species pools during restoration influenced beta diversity in the plant community at two scales: among prairies and within prairies (among plots).
  3. Among a group of prairies sown with larger species pools, among-site beta diversity was greater than among a group of prairies assembled from smaller pools, but not because of enhanced species sorting. We found an interaction between species pool size and an environmental filter, whereby beta diversity was higher among prairies restored with species-rich seed mixes, but only when soil moisture was also high. We detected neither greater beta diversity nor stronger species sorting among plots within prairies sown with species-rich mixes.
  4. Synthesis. This work provides what is to our knowledge the first large-scale manipulative test of how species pool size influences beta diversity. We found higher beta diversity among restored prairies sown with species-rich seed mixes, but little evidence for species sorting as a causal mechanism. Our results, based on manipulated real-world communities, provide an important link between previous theoretical and observational studies and small-scale experimental approaches. Of applied importance, our findings show that by creating communities of high beta diversity, ecological restoration can counteract widespread anthropogenic biotic homogenization.

Introduction

Why local communities differ from each other in composition and diversity is of longstanding interest in ecology. Central to this focus is the concept of beta diversity, the spatial component of species diversity which links local (alpha) and regional (gamma) diversity by describing variation in community composition across space. Since its formalization by Whittaker (1960, 1972), variation in beta diversity has been described across numerous gradients, including latitude, altitude and productivity (Soininen, Lennon & Hillebrand 2007; Harrison, Vellend & Damschen 2011; Mori et al. 2013). Understanding why some communities have low beta diversity and consistent community composition across space, while others have high beta diversity and high variability in the identity and abundance of species across localities, can provide insight into major determinants of community structure and diversity (Chase & Myers 2011; Kraft et al. 2011; Myers et al. 2013). From a pragmatic perspective, understanding the determinants of beta diversity will assist in addressing biotic homogenization caused by human-mediated global change (Olden 2006; Winter et al. 2009) through, for example, developing approaches to ecological restoration and land management that promote beta diversity (Matthews & Spyreas 2010). In spite of its basic and applied relevance, understanding the mechanisms underpinning patterns of beta diversity remains a significant challenge. In particular, we lack studies that manipulate the purported drivers of beta diversity across realistic spatial scales and under real-world field conditions. This has led to a lack of understanding of cause and effect relationships between gradients of hypothesized importance and the development of beta diversity.

The size of the species pool, the number of species that could potentially disperse to and persist at a site, is one potential determinant of beta diversity (Chase 2003; Fukami 2004; Harrison, Vellend & Damschen 2011; Kraft et al. 2011). Most empirical tests of this idea use observational gradients in regional (gamma) species richness as a surrogate for species pool size because of the difficulty of manipulating regional species pools. However, because regional species richness can result from local processes such as abiotic limits to species richness (Harrison & Cornell 2008) and from beta diversity itself (Qian et al. 2013), observational approaches cannot directly test for the role of species pool size. Disentangling the influence of species pools on beta diversity requires manipulative tests in systems where species pools are clearly defined and manipulated independently of the species present in the region.

Furthermore, moving from a description of pattern to a knowledge of process requires determination of whether a species pool effect is consistent across communities that vary in ecological context, as well as the ecological mechanism(s) by which species pool size underlies beta diversity. For example, harsh environmental conditions can reduce beta diversity by imposing filters that prevent establishment by species unable to tolerate these conditions (Chase 2007, 2010; Myers & Harms 2011). However, this effect may be mediated through species pool size because harsh environments reduce the number of species that can survive (Chase 2007; Germain et al. 2013). Therefore, while environmental filters should affect beta diversity, the effect may depend on species pool size.

Stronger species sorting provides one mechanism through which species pool size could lead to increased beta diversity (Questad & Foster 2008; Harrison, Vellend & Damschen 2011). Species sorting produces beta diversity along environmental gradients by enabling species to establish in environments to which they are well suited (Whittaker 1960; Leibold et al. 2004; Cottenie 2005); this has also been termed ‘structured beta diversity’ (Harrison, Vellend & Damschen 2011). Size of the species pool should strengthen species sorting, tightening the association between species and environment by increasing the likelihood that species are present at suitable points along environmental gradients (Questad & Foster 2008; Foster et al. 2011). This hypothesis remains poorly evaluated. Foster et al. (2011) showed that propagule addition can strengthen species sorting, but could not distinguish between effects of greater richness and effects of greater total abundance of seeds, which should also strengthen species sorting by reducing dispersal limitation and ensuring that species reach each point on the environmental gradient. It is essential to test for an effect of species pool size while controlling for variation in total seed density if we are to understand whether species sorting causes a positive effect of species pool size on beta diversity.

Beta diversity is an inherently spatial phenomenon that requires investigation at multiple spatial scales (Barton et al. 2013). Differences in beta diversity among community types may not be consistent at multiple spatial scales (e.g. Questad et al. 2011), and mechanisms increasing beta diversity at one scale may be weak or undetectable at another scale. However, we predict that species pool size should increase beta diversity at spatial scales both within a single continuous habitat patch and among spatially disparate patches. Species sorting along environmental gradients should be stronger in sites exposed to larger species pools, whether gradients and communities are assessed among plots within a patch or among patches. Furthermore, environmental harshness should increase compositional similarity at both scales. While our predictions are consistent across scales, testing the generality of these patterns across scales is a key step in understanding the determinants of beta diversity.

We tested the role of species pool size in shaping beta diversity across a series of restored prairies in southwestern Michigan. Ecological restoration presents unique opportunities to study community assembly (Temperton et al. 2004) because restored communities are independent assembly events. For constructed communities, we can define the species pool as the suite of plant species that were sown into bare soil at each site (Germain et al. 2013). In our landscape, natural sources of sown species are rare because of the widespread extirpation of native prairies and the relative rarity of restored prairies on the landscape. This affords the ability to link species pools to the resulting community of sown species. Beta diversity of these sown species can illustrate the roles of species pool size, environmental conditions and species sorting. Furthermore, restoration efforts provide manipulated settings where we can evaluate how assembly processes scale up to construct ‘real world’ ecological communities at large spatial scales and over relevant timescales.

For each of the 29 prairies in this study, we determined the species richness of the seed mix used for restoration (i.e. the size of the species pool) and the composition and richness of the sown community that established. We asked three specific questions at each of two spatial scales (within and among sites). (i) Does increased species pool size (seed mix species richness) increase prairie plant beta diversity? (ii) Does the effect depend on the strength of a key environmental filter (soil moisture)? (iii) Is the effect of species pool size on beta diversity caused by stronger species sorting?

Materials and methods

Study Sites

We sampled plant communities in 29 restored prairies spread across 4 counties (spanning an area of ~1300 km2) in southwestern Michigan, USA (see Fig. S1 in Supporting Information). All 29 sites had previously been tilled, and all sites were herbicided prior to restoration to prairie. Each site was seeded once in 2003–2008 with a mix of prairie plant species (Native Connections, Three Rivers MI). We confirmed that variation in site age did not affect our analyses (detailed below). For each site, we obtained data on seed mix species composition, species richness and evenness, and total seeding density (total mass of all seeds added per m2). Seed mix species richness ranged from 8 to 71, seeding density ranged from 0.7 to 1.3 g seed m−2 (97–186 oz ac−1) and seed mix evenness ranged from 0.23 to 0.48. More species-rich mixes had lower evenness (Spearman's ρ = −1.0). A few sites shared identical seed mixes, but most had unique mixes. Seed mixes were somewhat tailored to site environmental conditions (Fig. S2b) and variation in seed mix composition was clearly associated with variation in plant community composition across sites (Mantel r = 0.47, P = 0.001). However, community composition was not wholly pre-determined by seed mix composition, as the majority of variation in community composition was not associated with seed mix composition (Grman, Bassett & Brudvig 2013). Furthermore, we confirmed that while species-rich seed mixes were assembled from a longer list of candidate species (131 species) than the species-poor seed mixes (72 species), variation among seed mix composition was not greater among species-rich seed mixes compared to species-poor mixes (see Data analysis section for delineation of species-rich and species-poor seed mixes; Fig. S3). Therefore, if we find greater beta diversity in the plant communities of sites sown with high vs. low richness mixes, we can be confident that it is not driven by differences in beta diversity in the seed mixes.

Because seed mix assignment to sites was determined by individual landowners, we tested whether sites receiving species-rich or species-poor mixes differed systematically in a number of potentially important factors. Sites with species-rich mixes were not edaphically different in soil moisture (F1,27 = 0.18, P = 0.7) or a suite of other important soil variables including pH, soil organic matter, soil texture, total exchange capacity and Bray-I extractable phosphorus (F1,27 = 1.02, P = 0.4). They also did not differ in their frequencies of prescribed burning (F1,27 = 1.5, P = 0.2), their size (F1,27 = 3.78, P = 0.06; trend for low richness sites to be larger), or their immediately prior land-use histories (all sites were historically in tillage, but some were used as pasture or hay/old field prior to restoration). They did not have greater heterogeneity in soil moisture (Levene's test F1,27 = 0.04, P = 0.8), the suite of other soil properties (F1,27 = 1.75, P = 0.2), burn history (F1,27 = 1.29, P = 0.3) or size (F1,27 = 3.41, P = 0.08; trend for low richness sites to be more variable in area). Sites sown with species-rich seed mixes were not more geographically separated from each other than species-poor sites (Fig. S1). Finally, sites sown with species-rich mixes did not differ from sites sown with species-poor mixes in their surrounding landscapes (cover of grassland, agricultural land, forest, developed land or wetland; see Grman, Bassett & Brudvig 2013 for more information about landscape context). Therefore, any differences in beta diversity between sites sown with high and low richness mixes are likely a result of seed mix richness.

Plant Sampling

In August–September 2011, we sampled plant community composition along a randomly oriented 50-m transect in the centre of each site. In each of 10 1-m2 plots along the transect, we identified plants using Voss (1972, 1985, 1996) and visually estimated percent cover by each species. To test our hypotheses, we reduced the data set to presence–absence information because no existing ecological theory makes clear predictions about how species pool size should affect species relative abundances in the sites. To determine which species had been sown, we compared the list of observed species to the site's seed mix. Sown species dominated most communities, contributing 67 ± 22% of plant cover (mean ± SD). To evaluate relationships between the species pool and beta diversity of the plant community, we analysed only sown species. To estimate plant above-ground biomass, we clipped all plants (sown and non-sown) and dried and weighed the biomass.

Soil Sampling and Analysis

We took eight soil cores, 20 cm deep and 2 cm in diameter, evenly spaced around the perimeter of each 1 m2 plot. We sieved the soil through a 4 mm sieve, then air-dried the samples. We analysed the ability of sieved soil samples to retain moisture in the laboratory (hereafter ‘soil moisture’) as the proportional difference in saturated wet weight and oven-dried weight, following Brudvig & Damschen (2011). Soil moisture is a component of plant-available soil moisture, a critical environmental variable in grasslands, contributing to above-ground productivity, competitive interactions between plants, plant species distributions (Nelson & Anderson 1983; Knapp et al. 1998; Faber & Markham 2011), and a filter for assembling communities in Michigan grasslands (Houseman & Gross 2011). In our data set, above-ground plant biomass was positively related to soil moisture (r = 0.21, regression P < 0.001), indicating that soil moisture is likely an environmental factor that shapes community assembly in our sites.

Data Analysis

To measure the plot-to-plot variability in community composition within each site (within-site beta diversity), we calculated pairwise compositional dissimilarity among all 10 plots at each site (Anderson et al. 2011). We used Raup-Crick dissimilarity, an indicator of compositional differences based on species presence–absence data that are not sensitive to differences in species richness (Chase et al. 2011). We used the mean distance of each plot to the site centroid (mean dispersion around the centroid in principal coordinates space based on Raup-Crick dissimilarities; Anderson, Ellingsen & McArdle 2006) to indicate within-site beta diversity (Anderson et al. 2011). We used multiple linear regression to determine the influence of seed mix richness, site mean soil moisture, and their interaction on within-site beta diversity. To remove any potential effects of restoration site age (number of years since planting; Sluis 2002) and total seeding density (g seed m−2; Dickson & Busby 2009), we included those variables in the regressions; effects were all non-significant (> 0.2).

To test for species sorting along the gradient of soil moisture within each site, we conducted partial distance-based redundancy analysis (dbRDA) of Raup-Crick dissimilarities among plots at each site. This method of constrained ordination tests for community differentiation along specific environmental gradients (Legendre & Anderson 1999). In this analysis, we included each plot's position along the sampling transect as a conditioning variable to remove the effect of physical proximity and any unmeasured environmental gradients along the transect. We calculated the proportion of variation in community composition associated with the environment (the strength of species sorting) in each site by dividing the eigenvalue of the constrained axis (soil moisture) by the total inertia (Vellend et al. 2007). We calculated heterogeneity in soil moisture for each site as the coefficient of variation (CV; results were identical using standard deviations). We then used multiple regression to test whether the strength of within-site species sorting depended on seed mix richness, heterogeneity in soil moisture, or their interaction, again removing any potential effects of restoration site age and total seeding density (again those effects were non-significant; P > 0.8).

To assess whether species pool size and soil moisture increased compositional variability among sites (among-site beta diversity), we compiled data on species occurrences from all of the 10 plots at each site and used a test for homogeneity of multivariate dispersions (Anderson, Ellingsen & McArdle 2006). This analysis compares variability between discrete groups. To create groups, we divided sites into 2 approximately equal-sized groups based on their seed mix richness (high richness, >35 species in the seed mix, 14 sites; or low richness, <35 species in the mix, 15 sites). While our high-richness restorations were species-poor relative to native tallgrass prairie, they represent typical high-diversity restorations commonly employed in our study region. We then tested whether among-site beta diversity was greater among high mix richness sites than among low mix richness sites using the test for homogeneity of multivariate dispersions on Raup-Crick dissimilarities. We obtained qualitatively similar answers when we repeated the analysis with three groups and with four groups of prairies. In a similar analysis, we created two approximately equal-sized groups based on site soil moisture (high soil moisture, >0.4 g water per g soil, 15 sites; or low soil moisture, <0.4 g/g, 14 sites) and tested whether among-site beta diversity was greater among high soil moisture sites than low soil moisture sites. Our sites represent a large range of soil moisture (29.5–53.0); moisture was largely determined by clay content (which ranged from 4 to 20% among prairies), sand content (29–88%) and soil organic matter (1.2–7.9%; see Grman, Bassett & Brudvig 2013 for more detail). We visualized results using NMDS. To further ask whether seed mix richness and soil moisture interacted to determine among-site beta diversity, we split sites into four groups (high richness, high soil moisture, six sites; high richness, low soil moisture, eight sites; low richness, high soil moisture, nine sites; low richness, low soil moisture, six sites) using the same breakpoints as above. We compared among-site beta diversity in each of the four groups (pairwise comparisons of mean site dispersion around the group centroid; Anderson, Ellingsen & McArdle 2006). Beta diversity did not differ between the high-richness seed mixes applied to sites with high and low soil moisture (Fig. S4; P = 0.9), although at low seed mix richness, beta diversity among mixes was higher in low moisture sites (P < 0.001).

Finally, we investigated species sorting among sites. To understand whether communities differed among sites depending on site average soil moisture, seed mix richness or their interaction, we conducted partial dbRDA where each site was a data point. A significant interaction term would indicate that species sorting along the soil moisture gradient among sites differed in strength depending on seed mix richness. In this analysis, we included total seeding density and site age as conditioning variables to remove their effects before investigating the effects of seed mix richness, soil moisture and their interaction.

We conducted all analyses using R 2.15.3 (R Development Core Team 2013) packages vegan 2.0-6 (Oksanen et al. 2011) and ggplot2 0.9.3.1 (Wickham 2009), except for calculations of Raup-Crick dissimilarities where we used the implementation provided with Chase et al. (2011) for the ‘classic metric’ (to avoid negative dissimilarities) and split ties. We tested for the significance of predictors using marginal sums of squares and removed non-significant interactions. We examined correlations among predictor variables; seed mix species richness was correlated with total seeding density (r = 0.46) but all other correlations among predictors were <0.2. Importantly, a site's seed mix species richness was not correlated with its soil moisture (r = 0.08).

Results

Within-Site Beta Diversity and Species Sorting

Counter to our predictions, neither seed mix richness nor soil moisture independently affected within-site beta diversity (Fig. 1; F1,24 = 1.62, P = 0.2 and F1,24 = 0.71, P = 0.4, respectively). There was no evidence for an interaction between these two factors (F1,23 = 0.02, P = 0.9).

Figure 1.

Neither (a) the species richness of seed mix used at a site nor (b) the site mean soil moisture was associated with greater within-site beta diversity (multivariate dispersion in Raup-Crick dissimilarity among plots within a site). Partial coefficients of determination and P-values shown from multiple regressions.

We similarly found no effect of site seed mix species richness, heterogeneity in soil moisture, or their interaction on the strength of species sorting along the soil moisture gradient within each site (Fig. 2; F1,24 = 0.98, P = 0.3; F1,24 = 0.10, P = 0.8; and F1,23 = 0.28, P = 0.8). These results indicate that seed mix richness did not increase the strength of within-site species sorting.

Figure 2.

Within-site species sorting was not stronger with (a) a site's seed mix species richness or (b) a site's variability in soil moisture. Partial coefficients of determination and P-values shown from multiple regressions.

Among-Site Beta Diversity and Species Sorting

Seed mix richness increased among-site beta diversity (variability in community composition among sites; Fig. 3a; F1,27 = 7.98, P = 0.008). Seed mix richness also increased regional diversity: we found a total of 35 species across all 15 sites with species-poor mixes (<35 species in the seed mix), whereas we found 52 species across all 14 sites with rich mixes (>35 species in the seed mix). We had expected that sites with high soil moisture would have a weaker environmental filter and therefore develop greater among-site beta diversity, but we did not detect this (Fig. 3b; F1,27 = 1.19, P = 0.3). However, soil moisture interacted with seed mix richness to determine among-site beta diversity (Fig. 4; F3,25 = 10.10, P = 0.001). High soil moisture increased among-site beta diversity only when seed mix richness was high (P = 0.007), not when mix richness was low (P = 0.10). This was true in spite of equal numbers of total species observed (43) at high vs. low soil moisture sites, for sites restored with high mix richness. Mean site-level species richness also did not differ (17 species in low soil moisture sites and 18 in high). Together, these results suggest that soil moisture did indeed increase beta diversity, but only when seed mix richness was high enough to allow differences among communities to develop.

Figure 3.

Variability among sites in community composition (multivariate dispersion in Raup-Crick dissimilarity), our measure of among-site beta diversity, was smaller among (a) restoration sites seeded with a species-poor seed mix (<35 species; black dots; n = 15) than among sites seeded with a high richness seed mix (>35 species; white dots; n = 14), indicating more consistent community composition among sites with depauperate mixes. Among-site beta diversity was not greater among (b) sites with high soil moisture (white dots; n = 15) compared to sites with low soil moisture (black dots; n = 14).

Figure 4.

Low soil moisture (grey bars) restricted beta diversity (multivariate dispersion in Raup-Crick dissimilarity) among sites sown with rich seed mixes compared to high soil moisture (white bars), but not among sites sown with depauperate mixes (mean ± 1SE).

Our indicator of among-site species sorting, community differentiation among sites with different soil moistures, was not stronger in sites with species-rich mixes (interaction F1,23 = 1.20, P = 0.4), although site-level community composition differed with both site mean soil moisture and seed mix species richness (Fig. S2). Similarly, seed mix species richness did not alter the effect of other environmental variables on community composition, including pH (interaction F1,23 = 0.29, P > 0.9), Bray-I extractable phosphorus (F1,23 = 0.56, P = 0.8), the number of years since a prescribed fire (F1,23 = 1.08, P = 0.3) or immediately prior land-use history (F1,21 = 1.37, P = 0.2).

Discussion

Unravelling the determinants of beta diversity across latitudinal and environmental gradients may hold the key to understanding major drivers of community assembly, composition and diversity (Chase & Myers 2011; Kraft et al. 2011; Myers et al. 2013). Theory and empirical work suggest that species pool size can increase beta diversity (Chase 2003; Fukami 2004; Questad & Foster 2008; Kraft et al. 2011; Stegen et al. 2013), but this causative relationship has remained poorly investigated because of the difficulty of manipulating species pools across large spatial and temporal scales. Using a system of restored prairies for which the species pools (seed mixes) are known, we found mixed support for the hypothesis that species pool size increases beta diversity. Prairies assembled using larger species pools, that is, those sown with species-rich seed mixes (more than 35 species), supported greater beta diversity but only at large spatial scales (among restoration sites). This is despite equivalent beta diversity among high and low richness seed mixes (Fig. S3). We found no evidence for this effect at small spatial scales (within individual restoration sites), or for species sorting as a mechanism for the effect of species pool size on beta diversity among sites. We did, however, find support for a hypothesized interaction between an environmental filter (soil moisture) and species pool size, whereby prairies assembled with large species pools on harsher (low soil moisture) sites supported less beta diversity than did prairies on more benign (high soil moisture) sites. This pattern was not evident for prairies assembled with small species pools, where harsh and benign sites supported comparable beta diversity.

Our results advance previous observational and theoretical work (Chase 2003; Fukami 2004; Kraft et al. 2011; Myers et al. 2013; Stegen et al. 2013) because in our study, species pool size was not a consequence of local community composition or environmental conditions. In our study, sites with species-rich mixes did not differ in edaphic conditions, frequencies of prescribed burning, size, immediately prior land-use histories or landscape context. They did not have greater heterogeneity in soils and they were not more spread out geographically (see Materials and Methods for details). Most importantly, species pools were imposed on the sites by land managers, and were not a consequence of the local communities that established. This has an important bearing on our capacity to draw a cause and effect relationship between species pool size and beta diversity. Stegen et al. (2013) showed that environmental differences between regions of different species pool sizes can mask or amplify the apparent effect of pool size on beta diversity, emphasizing the need for studies where species pool size is manipulated independently of local conditions. In the only experimental study of which we are aware, Questad & Foster (2008) showed that beta diversity was greater in small-scale prairie plots sown with a high, relative to low, functional diversity of plant species (of equal species richness). Our study provides a key bridge between these past theoretical, large-scale observational and small-scale experimental studies by showing that species pool size enhances beta diversity across the large spatial scales that are relevant to the assembly of natural communities.

We expected that enhanced species sorting would mediate a positive relationship between species pool size and community diversity (Questad & Foster 2008; Foster et al. 2011; Stegen et al. 2013). We detected no evidence of this. Within sites, an effect of soil heterogeneity on the strength of community–soil relationships was not stronger in sites with species-rich mixes, as it would have been if species pool size had allowed for stronger species sorting. Among sites, associations between soil moisture and community composition did not strengthen with seed mix richness, suggesting that sites receiving high richness mixes did not assemble communities more suited to their environmental conditions than sites with low richness mixes. Seed mix richness also did not strengthen among-site species sorting along other environmental gradients (pH, soil organic matter, sand content, phosphorus availability, time since a prescribed fire or land-use history). While species could have sorted along unmeasured environmental gradients, it does not appear that the positive effect of seed mix species richness on among-site beta diversity was mediated through enhanced species sorting along any gradient we explored.

If not strengthened species sorting, what other mechanisms might have contributed to the effect of species pool size on among-site beta diversity? One potential explanation is that greater seed mix richness simply encouraged greater random differences in community composition (Hurtt & Pacala 1995; Fukami 2004; Kraft et al. 2011; Brownstein et al. 2012). There has been substantial recent interest in understanding the role of stochastic processes in community assembly (e.g. Chase & Myers 2011) and, while we provide no direct support, it is possible that large species pools enhanced the role of stochasticity during the assembly of the prairies we studied.

Second, in manipulating the richness of seed mixes applied to these restored prairies, land managers also adjusted the relative abundance of species in the mixes. Species-rich mixes are prohibitively expensive (thousands of dollars per ha) to landowners, largely because of the high cost of some species for which it is difficult to obtain seed. High richness mixes in this study included many of these difficult-to-obtain species in small quantities [0.5–2 oz/ac (<0.02 g m−2)]. Species sown in low densities were much less likely to establish (logistic regression of establishment on sowing density with species as a random factor, P < 0.001). Only 27% of species sown at <0.02 g m−2 established, while 94% of species sown at >0.1 g m−2 established. Large numbers of rare species in species-rich pools may partially explain why species-rich mixes assembled with greater beta diversity (Hurtt & Pacala 1995). However, we argue that this tight covariation of species richness and evenness in our species pools does not invalidate our study. In fact, this mechanism may also be at work in many natural systems, where large species pools frequently contain many rare species (Stirling & Wilsey 2001; Soininen, Passy & Hillebrand 2012). Whether species pool size (richness) per se or species pool evenness drives the effect of pool size deserves further theoretical and empirical investigation. Separating the influences of these stochastic drivers from deterministic drivers of beta diversity, such as species sorting, remains a major challenge for ecology.

Our results also support the hypothesis that beta diversity is influenced by an interaction between species pool size and environmental filters. Variation in community composition was higher among high soil moisture prairies sown with high (but not low) richness seed mixes. Previous work has suggested that effects of environmental conditions on beta diversity are mediated through effects on species pool size, such that there should be a smaller effective species pool in harsh (low soil moisture) sites because fewer species can survive those conditions (Harrison et al. 2006; Chase 2007, 2010; Lepori & Malmqvist 2009; Harrison, Vellend & Damschen 2011). However, this mechanism does not seem to be operating in our system; we observed the same number of species across both groups of sites. One possible alternative explanation is that high soil moisture enabled species sorting along some other gradient among sites, such as soil fertility or pH (Harrison, Vellend & Damschen 2011); our sample sizes were too small to reliably test for these three-way interactions. We suggest that future work should continue investigation of the mechanisms by which species pools affect beta diversity, including the potential for interaction with multiple gradients.

Our results show that prairie restorations sown with species-rich seed mixes exhibit higher among-site beta diversity. This has bearing on the prospects for ecological restoration to combat anthropogenic biotic homogenization. Restoration strategies that achieve a range of outcomes in terms of species composition can be important for maintaining desirable landscape heterogeneity (Matthews & Spyreas 2010). However, this finding also indicates that species-rich seed mixes result in more unpredictable prairie community compositions. Thus, sowing with a diverse mix may help land managers achieve one restoration goal, decreasing biotic homogenization across a suite of prairies, but concomitantly decrease their capacity to attain a second goal of a specific community composition in any particular site. More work is needed to advance prediction in restoration ecology (Brudvig 2011) and indeed to understand whether a single target community composition or range of acceptable outcomes (i.e. high beta diversity among restorations) is appropriate (Matthews & Spyreas 2010). Our work suggests that these efforts will need to confront trade-offs, such as the desire to increase diversity and achieve compositional predictability.

In one of the few studies where species pools are defined and manipulated independently of local communities, we confirm the importance of species pool size for enhancing beta diversity. Others have suggested that this relationship may play a crucial role in explaining patterns of greater beta diversity in, for example, high productivity, predator-free, or low-disturbance habitats (Harrison et al. 2006; Chase 2007, 2010; Chase et al. 2009) or in tropical forests (Kraft et al. 2011) where species pools are large. Testing the universality of the relationship across ecosystems, taxa and spatial scales is thus a critical step forward.

Acknowledgements

We thank the Edward Lowe Foundation, the Kalamazoo Nature Center, the Fetzer Institute's GilChrist retreat centre, the Southwest Michigan Land Conservancy, and private landowners for access to their prairie restorations. We also thank Jerry Stewart at Native Connections and Tyler Bassett for data on prairie plantings and site information. Tyler Bassett, Jonathan Landis, and Mike Epperly helped in the laboratory and field. Tyler Bassett, Ellen Damschen, Tim Dickson, Dani Fegan, Susan Magnoli, Xiangcheng Mi, Chris Steiner, Nate Swenson, and several anonymous reviewers provided useful comments on previous versions of the manuscript.

Ancillary