Fine-scale spatial heterogeneity and incoming seed diversity additively determine plant establishment


  • Paul J. Richardson,

    Corresponding author
    1. Centre for Ecosystem Resilience and Adaptation, Faculty of Environment, University of Waterloo, 200 University Avenue West, Waterloo, ON N2L 3G1, Canada
    2. The Ontario Aggregate Resources Corporation, Suite 103 1001 Champlain Avenue, Burlington, ON L7L 5Z4, Canada
    Search for more papers by this author
  • Andrew S. MacDougall,

    1. Department of Integrative Biology, University of Guelph, 50 Stone Road E., Guelph, ON N1G 2W1, Canada
    Search for more papers by this author
  • Douglas W. Larson

    1. Department of Integrative Biology, University of Guelph, 50 Stone Road E., Guelph, ON N1G 2W1, Canada
    Search for more papers by this author

Correspondence author. E-mail:


1. Plant establishment is critical for community assembly, but mechanisms regulating establishment can be obscured by covarying influences of incoming seed diversity and fine-scale spatial environmental heterogeneity (microhabitat heterogeneity). Both can maximize establishment, depending on whether species differences or environmental variability more fundamentally structures plant communities.

2. We experimentally assembled limestone-pavement herb communities to examine the relative effects of seed diversity and microhabitat heterogeneity on establishment. This included testing (i) whether effects of seed diversity strengthen with heterogeneity, as would be expected if potential niche differences are more strongly expressed in more heterogeneous environments and (ii) whether a greater number of incoming species can establish in more heterogeneous environments due to environmental filtering. Species interaction theory predicts that increased facilitation and niche complementarity with realized diversity has the potential to increase overall community density.

3. Heterogeneity operated independently of seed biodiversity and explained establishment even when spatially averaged microhabitat conditions were accounted for. Homogeneous plots sown with six species supported establishment of plant density that could be increased c. 10% by doubling added seed diversity (while holding heterogeneity constant), but increased c. 40% by holding seed diversity constant and maximizing heterogeneity.

4. Ordinations revealed that species establishment was sorted by gradients in soil pH and surface cover by moss, litter and open bedrock. Regressions indicated more species established in plots featuring a greater diversity of surface cover types and/or greater soil depth variability. Community density increased with established richness, and heterogeneity ceased to explain variance in density once established richness was included as an explanatory variable. All but one of the sown species exhibited increased population density with increased plot richness.

5.Synthesis: Community density in a high-stress environment increased with both fine-scale spatial heterogeneity and added seed diversity. However, these effects were independent of one another, and impacts of heterogeneity were stronger than those of seed diversity. Our results suggest heterogeneity promotes density indirectly, through downstream effects of enhanced establishment diversity such as facilitation. These findings confirm establishment-stage interrelationships among biodiversity, density and heterogeneity as overlooked determinants of community structure by providing important field support for ideas primarily tested in the greenhouse.


Short-term processes associated with plant establishment can have important long-term impacts on community structure (Foster & Tilman 2003; Houseman & Gross 2011), yet our understanding of factors controlling these processes is incomplete. While constraints posed by variation in light, soils, and biological interactions have been well studied for individual species (Harper 1961; Harper, Williams & Sagar 1965), requirements for establishment of multispecies communities from incoming seed remain unclear (Hutchings, John & Wijesinghe 2003). Efforts to restore or create new ecosystems within human-dominated landscapes are particularly challenged by this knowledge gap (Palmer, Ambrose & Poff 1997).

The diversity of incoming seeds and the degree of fine-scale spatial heterogeneity within a location can each regulate local establishment patterns through effects on density dependence and the outcome of species interactions. Biodiversity can regulate establishment, especially in homogeneous environments, if heterospecific neighbours occupy distinct niches and utilize resources in different ways (‘diversity model’) (Tilman 1994). Under these conditions, overall resource usage may be more efficient where more species establish, and such complementarity may promote a greater number of individuals. Similarly, diversity may influence establishment in high-stress environments by interspecific facilitation, as stress-ameliorating ‘benefactor’ and stress-sensitive ‘beneficiary’ species are more likely to co-occur and interact (Michalet et al. 2006). Alternatively, heterogeneity can regulate establishment when variation in fine-scale environmental conditions (the microhabitat) allows species to meet their diverse regeneration requirements (‘heterogeneity model’) (Ricklefs 1977; Gigon & Leutert 1996). Increased ‘patchiness’ of resources and stressors can provide increased refuge from limiting factors, enabling more species to experience ‘safe’ microhabitats for establishment (Grubb 1977). Even in single-species systems, spatial heterogeneity may promote establishment by reducing intraspecific competition, thereby reducing density-dependent constraints on population expansion (Rees, Grubb & Kelly 1996; Day, Hutchings & John 2003; Maestre & Reynolds 2006).

In addition to potential independent effects of seed diversity and microhabitat heterogeneity on establishment, theory suggests that at least two types of interactions may be important. On the one hand, if benefits of incoming seed biodiversity depend on different species occupying different niches, then factors promoting the expression of niche differences among neighbours may strengthen complementarity. Spatial heterogeneity may be one such factor, providing opportunities for species to fulfil functional roles that maximize their fitness (Tylianakis et al. 2008; Wacker et al. 2008; García-Palacios, Maestre & Gallardo 2011). On the other hand, if microhabitat variability promotes expression of niche differences crucial to establishment itself, a greater diversity of arriving species should establish in more heterogeneous locations due to species sorting along environmental gradients (‘ecological filtering’) (Questad & Foster 2008; Myers & Harms 2009; Foster et al. 2011; Pavoine et al. 2011). Initial establishment of more diverse communities may have far-reaching benefits for overall establishment of individuals (and associated ecosystem processes) due to facilitation or niche complementarity among species (Tilman, Reich & Knops 2006; Cardinale et al. 2007). Finally, these different mechanisms need not be mutually exclusive; heterogeneity may regulate establishment through effects on established diversity, but such effects may be more pronounced where incoming seed diversity is higher (Foster et al. 2011).

Here, we test ‘diversity’ and ‘heterogeneity’ models of plant establishment in limestone-pavement herb communities introduced to limestone quarry floors. Previous studies have rarely tested both models simultaneously and have been greenhouse experiments (e.g. Maestre, Bradford & Reynolds 2006). The rock barrens studied here are ideal for investigating how heterogeneity and seed diversity regulate establishment due to significant variation in the degree of heterogeneity existing at microhabitat scales. We manipulated seed diversity in small-plot communities by adding seeds of 3, 6 or 12 species to vegetation-free patches of quarry floor, and we replicated seed diversity treatments extensively to capture a wide range of variation in microhabitat conditions. We investigated how the number of plants established over a 2-year period depended on the diversity of added seeds and the heterogeneity of the microhabitats into which these seeds were sown (Lundholm 2010).

We predicted that if heterogeneity is important, then the overall density of individuals established would be greater in plots featuring more spatially variable microhabitats. If seed biodiversity is important, then y-intercepts for regression relationships between heterogeneity (x) and established density (y) should be greater in plots sown with more species. However, there may also be interactive effects. Where effects of seed diversity strengthen with heterogeneity, slopes of heterogeneity–density relationships should become steeper with increased seed diversity. With respect to mechanism, if heterogeneity primarily influences density through indirect effects of established biodiversity (such as increased facilitation), then established richness should increase with heterogeneity while density increases with established richness. However, if increased density with established richness is primarily due to a few species with higher-than-average density (e.g. smaller-than-average plants) occurring more frequently in richer plots, then only a few small-statured species should exhibit increased population density with plot richness. Alternatively, if generally positive interactions among multiple species underlie increased community density with richness, then most species should exhibit increased population density with plot richness (Richardson, Horrocks & Larson 2010).

Materials and methods

Study system

Our experiment was conducted at two dry limestone quarries in southern Ontario, Canada that were abandoned without rehabilitation 35 and 20 years prior to the study (Waters quarry: 43°41′16.9″ N, 79°57′33.5″ W; Fischer quarry: 43°24′55.0″ N, 80°07′55.6″ W). Experimental sites were characterized by high ‘patchiness’ of microhabitats, ranging from open bedrock to partially shaded soil-filled crevices, and these supported diverse but spatially sporadic low-growing vegetation including stress-tolerant forbs, graminoids, shrubs and bryophytes. Mean soil depth was approximately 4 cm but highly variable, ranging from 0 cm to more than 30 cm. Quarry substrates were alkaline (mean pH = 7.8), of low to moderate fertility, and contained a mineral fraction of crushed CaCO3 residual from quarrying (for further details on soil properties, see Tomlinson et al. 2008).

As these features also characterize naturally occurring alvar limestone pavements, which are biodiversity hotspots in Canada and elsewhere (Schaefer & Larson 1997; Znamenskiy, Helm & Partel 2006), abandoned quarries are increasingly being restored with native alvar vegetation (Matthes et al. 2009). Strategies aimed at establishing such vegetation rely on seed addition of alvar assemblages to patches of quarry floor, but restoration success is influenced by the fact that even within relatively small patches, cover by specific microhabitat types can range from homogeneous to highly heterogeneous (Richardson, Lundholm & Larson 2010).

Substrate amendment

At each site, 100 (0.3 × 0.3 m) plots were installed at random locations on the quarry floor. Glyphosate herbicide (Round-Up®; Monsanto, Creve Coeur, MO, USA) was applied to each plot prior to seed addition, to kill resident vegetation and minimize potential impacts of these plants on interactions among the focal (sown) alvar species. The applied context of our study dictated that alvar seeds not be ‘wasted’ on patches of completely barren pavement where establishment is unlikely, given that the target community flourishes in the shallow-soil cracks and depressions that punctuate alvar pavements, rather than on the open pavement flats themselves. We anticipated that extreme stress imposed by complete absence of soil in some locations could overwhelm and obscure potential heterogeneity effects in the less-stressful patches, which serve as starting points for quarry restoration. For these reasons, we added to all plots a thin layer of manufactured substrate consistent with quarry-to-alvar restoration protocols (Larson et al. 2006). Rocks, gravel, herbaceous litter, woody debris, moss cushions and other non-soil microhabitat features were maintained in plots to the greatest extent possible by removing individual features immediately before substrate addition and replacing them immediately afterwards.

To each plot, we added 1.25 L of substrate comprised of horticultural-grade silica sand (Nu Gro, Brantford, ON, Canada) mixed equally (by fresh weight) with recycled agricultural compost previously used for mushroom production (‘PC Mushroom Compost’, President’s Choice, Toronto, Canada). Mushroom compost is generally comprised of decomposed, steam-sterilized straw and animal manure mixed at a ratio of c. 95 : 5 by fresh weight (Fidanza et al. 2010). The added mixture of compost and sand was 0.55% N, 0.14% P, 0.52% K and 13% organic matter by weight, values that are intermediate between typical quarry soils and natural alvar soils (Tomlinson et al. 2008). Bulk density of the added substrate was c. 0.8 g cm−3, equivalent to that of both natural alvars and quarries. Substrate addition initially increased soil depth by c. 1 cm.

We anticipated that applying substrate consistently within and among plots would not substantially alter among-plot differences in microhabitat heterogeneity. We tested this assumption by comparing environmental data from the study plots against previously collected data from similar plots that did not receive substrate and that were located on patches of quarry floor containing some soil naturally. This comparison indicated that quarry plots that received soil and those that did not were statistically indistinguishable with respect to all tested measures of microhabitat heterogeneity, and most metrics of spatially averaged plot conditions (Appendix S1 in Supporting Information).

Seed addition

Each plot received 1200 viable seeds (i.e. a seeding rate of c. 13 000 seeds m−2) with the number and identities of seeds varying according to five seed diversity treatments. Both quarries and alvars are associated with regional species pools comprised of 180–200 herbaceous species (Tomlinson et al. 2008), but we selected 13 species to introduce experimentally based on confinement to alvar habitat (Catling 1995), seed availability and successful establishment during a previous study in this system (Richardson, Lundholm & Larson 2010). Non-target plants that spontaneously colonized plots were weeded out every 2–4 weeks over the growing season.

Seed diversity treatments were designed to: (i) isolate effects attributable to more rather than just different species in higher-diversity mixtures; (ii) consider a single gradient corresponding to both species richness and underlying trait diversity (Naeem 2002) and (iii) permit high replication, so that each mixture could be applied to multiple locations varying randomly in their degree of fine-scale spatial heterogeneity. The maximum-diversity treatment (‘12’) consisted of a pool of 12 species drawn from three functional groups (three sp. spring-flowering forbs; five sp. summer-flowering forbs; four sp. perennial grasses). Two intermediate-diversity treatments were derived by partitioning ‘12’ into two non-overlapping pools that each consisted of six species from two functional groups (‘6a’: three sp. spring forbs + three sp. summer forbs; and ‘6b’: four sp. grasses + two sp. summer forbs). Two low-diversity treatments were derived by partitioning ‘6a’ into two non-overlapping pools that each consisted of three species from a single functional group (‘3a’: three sp. spring forbs; and ‘3b’: three sp. summer forbs). Effects of seed diversity (isolated from composition) were thus detectable as performance differences between either: (i) ‘6a’ and the combination of ‘3a’ and ‘3b’; or (ii) ‘12’ and the combination of ‘6a’ and ‘6b’, given that each species had equal opportunity to establish at each of the contrasted levels. This design expands on other recent alternatives to random-draw/additive partitioning approaches to biodiversity experiments (Benedetti-Cecchi 2004; Bell et al. 2009). Each seed diversity treatment was applied to 20 replicate plots per site. The species composition for each treatment is shown in Table S2 (Appendix S3).

Although our design initially called for a total pool of 12 species, shortly prior to planting, it became apparent that one species in the summer-forb group was not sufficiently available to sow all plots at the required density. For this reason, half of the plots requiring Rosa blanda instead received Asclepias syriaca. These subgroups did not perform differently in any respect (data not shown) and thus are not distinguished further.


We assessed community establishment at the end of the first growing season and once mid-way through the second, each time assessing three main properties of the sown community: (i) the total number of individuals per plot (established density); (ii) the total number of species per plot (established richness) and (iii) the identities and abundances of these species (established composition, based on the number of conspecific individuals per plot (population density), for each sown species).


Microhabitat conditions in plots were measured midway through the first growing season, with spatial subsampling used to calculate both the spatial mean and variability of frequencies. Each plot was divided into 36 (5 × 5 cm) subplots, and at the centre of each subplot, soil depth was measured, and the surface contacted by the depth probe was classified as one of nine possible microhabitat types: surface cover by exposed bedrock, till, gravel, bare soil, woody debris, litter, bryophyte biomass, lichen biomass or algal/bacterial biomass (for details, see Lundholm & Larson 2003). Soil cores were collected from five randomly selected subplots per plot, using a 2-cm-diameter corer inserted to a maximum depth of 10 cm. These samples were analysed in the laboratory for substrate pH, gravimetric water content and bulk density. For each soil property, arithmetic means of subplot measures estimated ‘average’ plot conditions, while coefficients of variation (CV; the standard deviation divided by the mean) estimated spatial heterogeneity. ‘Average’ microhabitat composition in each plot was estimated for each microhabitat type as the percentage of subplots covered by that microhabitat type. Microhabitat heterogeneity was further calculated using the Shannon diversity index, accounting for the number and relative abundances of different microhabitat types (rather than different species) in each plot (‘Microhabitat H’).


Using mixed linear models (Proc Mixed and Proc GLM in sas v. 9.1; The SAS Institute, Cary, NC, USA), we analysed established density (log-transformed to normalize residuals and meet assumptions of linear models) as a function of seed diversity treatment, microhabitat heterogeneity and ‘average’ microhabitat conditions. This model-building approach is analogous to multiple regression, requiring the same caution regarding making inferences about the causal importance of variables that were not experimentally manipulated. Sampling year and site location were treated as random effects, with plots treated as the subject of repeated measures. Non-significant (i.e. > 0.05) variables were sequentially removed (based on highest P-value) from initial models that incorporated the full suite of microhabitat variables. Potential interactions between significant microhabitat variables and seed diversity were assessed, and significant interactions were retained in the final model. Intercepts and slopes of microhabitat establishment–regression relationships were contrasted between diversity treatments differing in sown richness, but not overall composition (i.e. ‘12’ vs. the average of ‘6a’ and ‘6b’ combined; ‘6a’ vs. the average of ‘3a’ and ‘3b’ combined). For heterogeneity–density relationships, intercept differences quantify establishment benefits gained by doubling seed diversity in a homogeneous microenvironment. Slope differences quantify how such benefits change as microhabitat heterogeneity increases.

Established composition of sown communities was analysed using multivariate ordination analyses to test whether species sorted along environmental gradients. Using canoco v. 4.5 (ter Braak & Smilauer 2002), we first performed detrended correspondence analysis (DCA) to assess the pattern of compositional variance among plots regardless of treatment and environmental variables. Partial canonical correspondence analysis (pCCA) was then conducted to assess how microhabitat variables explained residual variance once effects of treatments, site location and sampling year were accounted for as known sources of variance. All metrics of ‘average’ microhabitat conditions were tested as explanatory variables using 500-permutation Monte Carlo analysis. Non-significant variables were removed sequentially, and the final model was used to create species ordinations weighted by microhabitat variables (using Hill’s scaling, with rare species down-weighted). We expected that if different species had different microhabitat preferences or tolerances, then microhabitat gradients would explain significant variance in established composition.

To test whether heterogeneity potentially influenced established density through impacts on established richness, we used linear model analysis to evaluate: (i) the response of established richness (log-transformed) to microhabitat variables and (ii) the response of established density to established richness and microhabitat covariables. We expected that not only would established richness respond strongly to heterogeneity, but also established density would respond more strongly to richness than to heterogeneity, and the influence of heterogeneity on density would weaken upon inclusion of established richness as an explanatory variable.

Given the potential sensitivity of our results to self-thinning and other dynamics associated with the earliest stages of community assembly, we conducted an alternative to the repeated-measures analysis, to test whether early developments influenced community properties over the longer term. Specifically, we evaluated the response of plot density as established at the end of the second growth season to plot richness as established at the end of the first growth season. The same covariables were included in this analysis as in the main (repeated-measures) analysis.

We evaluated the performance of each individual species to test whether increases in community density with established richness result from many vs. few species increasing in population density. We thus analysed population densities of each species in each plot (where sown) in response to species identity, established plot richness, experimental and environmental variables and relevant interactions. Populations nested within plots were treated as the subject of repeated measures. Regression coefficients for the response of population density to plot richness were derived for each species from inspection of the interaction between species identity and established richness. As precise measures of plant size were not available for the species used, we tested whether smaller-than-average species showed larger-than-average increases in population density with plot richness by ranking the species according to visually estimated above-ground cover by 1- to 2-year-old individuals, and graphically inspecting the relationship between coefficient magnitude and rank species size.


Response of establishment to heterogeneity and seed biodiversity

The number of sown plants to establish in plots was highly variable, ranging from 0 to 251 in the first year and from 0 to 115 in the second. The back-transformed 95% confidence interval for least-squared mean density in year one was 13–18 individuals per plot, while in year two it was 10–14 individuals per plot. Although year was a significant predictor of density (Table S2 in Appendix S3), it did not interact with other factors, indicating that variables related to the experimental design or the microhabitat had constant effects over time. Density depended significantly on the seed diversity treatment and on several microhabitat variables that estimated either spatial averages or spatial variability of environmental features. Seed diversity alone explained 37% of the variance in density while ‘average’ microhabitat variables explained 7% of the variance (or 15% when interactions with seed diversity were included). ‘Variability’ variables explained 16% of this variance (or 18% when interactions with seed diversity were included; Table S2 in Appendix S3). Density increased with mean soil pH and decreased with both surface cover by till and mean soil moisture, but moisture and pH effects were only significant in plots sown with 12 species (Fig. S2 in Appendix S3).

Above and beyond these effects of ‘average’ conditions, established density responded strongly and positively to two metrics of within-plot spatial heterogeneity: Microhabitat H’ and soil depth CV (Fig. 1). Neither heterogeneity effect interacted with the seed biodiversity treatment. Although soil pH CV was also significant (Table S2 in Appendix S3), inspection of the significant interaction of this variable with seed diversity indicated the positive effect of heterogeneity was confined to a single seed diversity treatment (‘3a’).

Figure 1.

 Established plant density as a function of incoming seed diversity and two types of microhabitat heterogeneity. Left-hand panels show the response of density to Microhabitat H’, a diversity index reflecting both the richness and evenness of different microhabitat types present within a plot (regression coefficient = 2.8 ± 0.4, T112 = 6.26, < 0.0001). Right-hand panels show the response of density to the coefficient of variation among 18 spatial subsamples of soil depth taken per plot (regression coefficient = 1.3 ± 0.3, T112 = 4.51, < 0.0001). Slopes of regression lines in each panel are significant, but not significantly different from each other. Y-intercept values (‘b’) for regression slopes in the same panel significantly differ from one another where indicated (***< 0.0001), but regression slopes (‘m’) did not differ among seed diversity treatments.

Under average environmental conditions, plots seeded with either three-species mixture performed equivalently to plots receiving all six species combined, but plots seeded with 12 species exhibited 11% greater density than plots seeded with either six-species mixture (contrast of difference: F1,112 = 16.94, < 0.0001). By comparison, plots seeded with six species exhibited a 40% increase in density across the Microhabitat H’ gradient and exhibited a 37% increase in density across the soil depth CV gradient (Fig. 1). The relative advantage of a plot experiencing increased heterogeneity vs. increased seed biodiversity was estimated by contrasting predicted density in maximally heterogeneous plots receiving six species against that in maximally homogeneous plots receiving 12 species. For Microhabitat H’, heterogeneous six-species plots exhibited 23% greater density than homogeneous 12-species plots (contrast of difference: F1,112 = 12.17, = 0.0007). For soil depth CV, heterogeneous six-species plots exhibited 9% greater density than homogeneous 12-species plots (contrast of difference: F1,112 = 3.34, = 0.07). However, as seed diversity and microhabitat heterogeneity effects were positive and additive, maximum density occurred in heterogeneous plots receiving the 12-species seed mixture (Fig. 1).

Established richness varied strongly, ranging from 0 to 12 species plot−1 each year and averaging 2.0–2.3 species per plot (back-transformed 95% confidence interval for the least-squared mean over both years, as year was not a significant predictor of richness). This variance was mostly explained by the seed biodiversity treatment in combination with seven different measures of ‘average’ microhabitat conditions and three estimators of microhabitat heterogeneity (Table S3 in Appendix S4). Seed diversity alone explained 28% of the variance in established richness, while ‘average’ microhabitat variables explained ∼13% of the variance (24% including interactions with seed diversity), and heterogeneity variables explained 16% of the variance (18% including interactions with seed diversity; Table S3 in Appendix S4). Established richness decreased with mean soil moisture, but increased with mean soil pH, total surface cover by bare soil and cover by gravel (Fig. S2 in Appendix S3). The effect of pH only manifested in plots sown with 12 species, however, while effects of soil cover were stronger in plots sown with six rather than 12 species (Table S3 in Appendix S4).

Established richness increased significantly with Microhabitat H’, soil depth CV and soil pH CV (Fig. 2, Table S3 in Appendix S4). The response to soil pH CV was marginally weaker in plots receiving the three species than in plots receiving six (= 0.08); otherwise, all influences of heterogeneity on established richness were independent of sown richness (Table S3 in Appendix S4; Fig. 2). In general, plots sown with more species established more species, but the full complement of sown species rarely established in any single plot, regardless of seed mixture.

Figure 2.

 Established plant richness in response to incoming seed diversity and three types of microhabitat heterogeneity. Left-hand panels show the response of richness to Microhabitat H’ (regression coefficient = 1.2 ± 0.4, T92 = 3.15, = 0.0022). Centre panels show the response of richness to the coefficient of variation among 18 spatial subsamples of soil depth taken per plot (regression coefficient = 0.7 ± 0.2, T92 = 3.82, = 0.0002). Right-hand panels show the response of richness to the coefficient of variation among five spatial subsamples of soil pH taken per plot (regression coefficient = 0.12 ± 0.06, T92 = 2.07, = 0.0417 for plots sown with species group ‘6a’, and 0.10 ± 0.04, T92 = 2.42, = 0.0176 for the average of plots sown with either group ‘6a’ or ‘6b’). Significant differences between diversity groups with respect to regression intercepts (‘b’) and slopes (‘m’) are indicated (***< 0.0001; †0.05 < < 0.10).

Response of community composition to environmental factors

Detrended correspondence analysis yielded large eigenvalues (e.g. λ1 = 0.726 and λ2 = 0.152, cumulatively explaining 40% of variance in established composition), indicating strong patterning in the compositional differences among plots. Eigenvalues from pCCA were considerably smaller (e.g. λ1 = 0.019 and λ2 = 0.016, cumulatively explaining only 3.7% of species data, but 71% of species–environment relationships). This is consistent with most patterning in composition arising from the seed diversity treatment, which was controlled for as a source of known variance in pCCA. However, Monte Carlo analysis indicated that compositional variance that was not attributable to the seed biodiversity treatment was significantly explained by variation in mean soil pH (λA = 0.01, F = 2.85, = 0.004) and surface cover by bare rock (λA = 0.01, F = 2.15, = 0.021), litter (λA = 0.01, F = 3.21, = 0.001) and moss (λA = 0.02, F = 2.98, = 0.004; Fig. 3).

Figure 3.

 Partial canonical correspondence analysis (pCCA) of community composition in quarry floor plots sown with alvar herb species. Each species is labelled with a four-letter code corresponding to the genus name (Table S1) and symbolized by circles for late-flowering forbs, triangles for early-flowering forbs and squares for grasses. Distance between symbols in ordination space corresponds to decreasing likelihood that represented species will establish in the same plot. Plotted eigenvectors correspond to measured environmental variation that significantly influenced species occurrence (i.e. < 0.05 in Monte Carlo analysis). Species symbols separated from one another along the direction of an eigenvector have peak abundances at different points along the environmental gradient. The seed biodiversity treatment and sampling year were treated as known sources of variance (covariables), and thus patterns shown illustrate establishment patterns independent of these factors. Eigenvalues (λ) for the axes shown are λ = 0.019 for pCCA Axis 1 and λ = 0.016 for pCCA Axis 2.

Response of established density to established richness and microhabitat

When established density was evaluated as a function of established richness and microhabitat factors, heterogeneity variables became non-significant, regardless of the order by which terms were entered into the model. The best linear model explained 87% of the variance in density, with significant predictors including seed diversity (VE = 37%), established richness (VE = 41%) and seven metrics of ‘average’ microhabitat conditions (VE = 9%; Table S4 in Appendix S5). Community density increased strongly and monotonically with established richness, and to the same extent in each of the contrasted seed diversity groups (Fig. 4). Environmental variables had constant effects among the seed diversity groups, except for mean soil depth, which was associated with increased density in plots receiving 3 or 6, but not 12 species (Table S4 in Appendix S5). Analysis of year 2 density as a function of year 1 established richness (plus all relevant covariables) also indicated a strong, positive, monotonic response of density to richness (Table S5, Fig. S4 in Appendix S5). Year 1 richness explained 67% of the variance in year 2 density, and log density increased at a rate of 1.2 ± 0.1 individual plants per plot for each additional (log-transformed) species established (Fig. S4 in Appendix S5). This relationship was independent of the seed biodiversity treatment (Table S5 in Appendix S5).

Figure 4.

 Established plant density in response to incoming seed diversity and the richness of sown species to establish. All regressions were highly significant, but slopes and intercepts in the same panel were not significantly different from one another. Regression coefficients for the contrasted seed diversity groups were as follows: (three sp. avg.) 1.3 ± 0.1, T129 = 12.55, < 0.0001; (‘6a’) 1.2 ± 0.1, T129 = 9.37, < 0.0001; (‘6b’) 1.5 ± 0.1, T129 = 15.55, < 0.0001; (‘12’) 1.3 ± 0.1, T129 = 10.35, < 0.0001).

Response of population density to plot richness and species identity

Variance in established population densities of sown species was primarily explained by interactions between the identities of the species and plot properties including established richness and seven ‘average’ microhabitat variables (mean soil depth; total cover by rock, till, gravel, soil, litter and moss; Table S6 in Appendix S6). The interaction between established richness and species identity explained 47% of the variance in population density while the remaining environment-by-species interactions collectively explained c. 9%. An additional 1% of variance was explained by year of sampling, but year did not interact with other factors. Species differed with respect to their responses to established richness, but only one species (A. syriaca) failed to exhibit significantly increased population density with plot richness (Fig. S5 in Appendix S6). Three species exhibited relatively strong positive responses to richness, with regression coefficients ranging from 0.8 to 1.0 (±0.1), while the remainder of the species exhibited moderate positive responses with regression coefficients ranging from 0.18–0.30 (±0.06). Graphical inspection of these coefficients arranged according to species size indicated that while a few mid-sized species responded strongly to increased richness, species of all sizes increased similarly in density with increased richness (Fig. S5 in Appendix S6).


Constraints on plant establishment imposed by both seed and microhabitat limitations are important determinants of community structure (Turnbull, Crawley & Rees 2000; Zobel et al. 2000; Foster et al. 2004), and variability in species arrival and microhabitat availability may each potentially regulate communities through several mechanisms (Maestre & Reynolds 2007; Foster et al. 2011). Here, we found both types of variability to be important, but impacts of microhabitat heterogeneity were substantially stronger than those of added seed diversity. More added species meant greater establishment of individuals, but density was greatest where added diversity translated to established diversity most efficiently, which was within heterogeneous plots. In homogeneous environments, some species won over others early-on, thereby reducing diversity and its downstream benefits for individuals by facilitation. Plant communities were thus regulated by differences among incoming species interacting with heterogeneity, which permitted expression of species differences at the establishment stage via species sorting. Our study demonstrates that maximum expression of incoming biodiversity can depend importantly on fine-scale spatial diversity of the underlying environment.

In the high-stress system studied here, we found that plant establishment derived from both the diversity of species sown and the microhabitat into which they were sown, even when we controlled the effects of species and microhabitat identity. Heterogeneity had a stronger influence than seed diversity, and effects were additive such that established density peaked in heterogeneous microhabitats receiving the maximum diversity of seeds. Species sorted along environmental gradients and a greater number of sown species established in more heterogeneous plots, supporting heterogeneity models of community assembly (Ricklefs 1977). However, more individuals established in plots where more species established, supporting diversity models (Tilman 1994). Community density thus may have increased in heterogeneous environments as a result of strengthened facilitation or complementarity where more species occurred together.

Response of establishment to seed biodiversity and heterogeneity

The density of established alvar herbs responded positively to within-plot spatial variability of soil depth and to the combined richness and evenness of available microhabitat types (Microhabitat H’). These heterogeneity metrics explained density when the effects of mean soil depth and microhabitat cover were accounted for, indicating heterogeneity effects were not artefacts of correlations between spatial variability and average environmental conditions that contribute to community regulation (Stevens & Carson 2002; Lundholm 2010). Importantly, heterogeneity had the same positive effects on density regardless of incoming seed diversity, illustrated by the equivalent responses to heterogeneity among contrasted seed diversity groups. Significant differences in the intercepts but not the slopes of these heterogeneity–density relationships among 6- and 12-species seed diversity groups indicated an overall positive effect of seed diversity on density, but lack of difference between 3- and 6-species groups suggests a moderate-diversity threshold had to be passed before positive effects emerged. Plots receiving 12 species established 11% greater density than those receiving six species. By comparison, the c. 40% increase in density along microhabitat heterogeneity gradients suggests heterogeneity is a more important regulator of establishment than seed diversity in this system. Density was 23% greater in scenarios where microhabitat diversity was maximized and seed diversity held constant than where heterogeneity was held constant and seed diversity maximized.

Although spatially averaged conditions were mainly investigated to account for potential covariates of heterogeneity, different responses to these conditions among seed diversity levels revealed some interesting patterns. While it is unsurprising that high average soil moisture suppressed establishment while alkalinity promoted it (locations of maximum soil moisture were ephemerally flooded, and sown species were generally calcicoles), it is intriguing that plots sown with 12 species responded much more strongly to these variables than plots sown with fewer species (Fig. S2 and S3 in Appendices S3 and S4). All levels of seed diversity established poorly in the wettest locations, but plots sown with 12 species responded more positively to dry soils than did plots sown with six species. This could be due to increased importance of diversity-enhanced water use efficiency where drought was more limiting (Mulder, Uliassi & Doak 2001; Kahmen, Perner & Buchmann 2005). In contrast, plots sown with 12 species may have responded so positively to soil pH as a result of complementary usage of nutrients that were not biologically available in more acidic soils (Brady 1974).

A potential mechanism for heterogeneity’s impact on plant density?

Ordinations showed that variation in species composition not attributable to seed diversity was explained by variation in mean soil pH and cover by litter, moss and bare rock, suggesting sown species differed in their microhabitat preferences (Vivian-Smith 1997; Questad & Foster 2008). Linear model analysis indicated that more species established in plots exhibiting greater microhabitat diversity and spatial variability in soil depth and pH, as well as in plots with relatively dry soils and high overall cover by gravel or soil. Heterogeneity effects on established richness were independent of seed diversity. Taken together, these patterns suggest microhabitat heterogeneity promoted established richness by supporting a greater diversity of species regeneration requirements and increasing the degree of species sorting possible within plots (Questad & Foster 2008). Given the severity of the environment, the short term over which effects emerged, and the life stages investigated, it seems more likely that heterogeneity increased availability of diverse stress-refugia rather than disrupted competition, but further experiments must test this. Positive effects of spatial heterogeneity on species richness have been observed (Rey Benayas, Scheiner & Franklin 2002; Pausas et al. 2003; Dufour et al. 2006) or demonstrated experimentally (Vivian-Smith 1997; Gundale et al. 2006; Questad & Foster 2008) in several systems, including limestone pavements (Lundholm & Larson 2003). However, neutral and negative relationships have been found just as frequently (Grime et al. 1987; Stevens & Carson 2002; Baer et al. 2004; Reynolds et al. 2007). Future studies may resolve such discrepancies by differentiating between stress- and competition-reducing impacts of heterogeneity, and assessing the contingency of these upon environmental severity and life stages considered.

In addition to more sown species establishing in more heterogeneous microhabitats, more individuals established where more species established. Although heterogeneity initially explained 18% of the variance in density over 2 years, it ceased to be significant when established richness over 2 years was included as a predictor variable (which explained 41% of the variance in density). Seed diversity explained less variance in community density than heterogeneity, and established richness affected density equivalently in the contrasted seed diversity groups (Fig. 4). The positive response of density to richness was even more pronounced when density established in the second year was analysed as a function of richness established in the first year (Fig. S4 in Appendix S5). These patterns are consistent with an indirect impact of heterogeneity on density, mediated by effects of established richness (which is directly promoted by heterogeneity). As plot richness correlated positively with population densities for every sown species but one, many individual species likely interacted more positively where more species established, contributing to increased community density more or less equivalently. The species that responded especially strongly to plot richness were comprised of mid-sized rather than small plants (Fig. S5 in Appendix S6), which is inconsistent with the possibility that increased community density with richness arose from smaller, denser-growing species establishing preferentially in more species-rich plots.

The mechanism proposed here differs from the idea that heterogeneity strengthens niche complementarity by enhancing expression of niche differences (Tylianakis et al. 2008); however, the difference is mainly one of life stage considered. Whereas increased expression of niche differences among adults involves divergent growth or resource allocation patterns that may promote niche complementarity, increased expression of niche differences among arriving seeds involves differences in germination, emergence and survival – amounting to ecological filtering where niche differences relate to environmental differences (Grubb 1977; Myers & Harms 2009; Pavoine et al. 2011). Enhanced density from richness effects that themselves arose through species sorting is thus essentially an outcome of complementarity occurring within the domain of the regeneration niche. However, as we did not control established richness experimentally, further work is needed to demonstrate causation of this mechanism.

Conclusions and synthesis

Our findings have three fundamental implications for understanding controls of community membership and processes, in addition to applications for ecosystem rehabilitation (discussed in Appendix S7). First, spatial variability at the scale of the microenvironment can increase community density and species richness during establishment. While longer-term monitoring will ultimately determine how long-lasting these effects are, filters on establishment processes are recognized as critical to community structure over the long term (Foster & Tilman 2003; Foster et al. 2011). Second, incoming seed diversity has positive effects on community density that are independent of microhabitat heterogeneity, such that seed diversity and heterogeneity additively promote establishment of individuals. Third, the mechanism by which heterogeneity promotes density appears to be indirect, mediated by positive effects of heterogeneity on established richness (via ecological filtering) and positive effects of established richness on density (via facilitation or niche complementarity). The independence of seed and microhabitat diversity effects superficially contradicts predictions that biodiversity effects strengthen with heterogeneity due to increased opportunity for niche complementarity. However, our results are more consistent with this hypothesis when the role of heterogeneity-dependent established richness is considered. We propose that, in this system, heterogeneity increased expression of regeneration niche differences that ultimately resulted in ecological filtering, increased established richness and consequent promotion of community density. Continuing to improve our understanding of such dynamics as they occur in real-world ecosystems (rather than in experimental greenhouses) will thus be critical in developing effective applications for different types of ecological diversity in conservation and restoration management.


This work was funded by the National Science and Research Council (NSERC) and The Ontario Aggregate Resource Corporation. Thanks to L. Klatt, M. Bell, T. Wilson, K. Kuntz, M. Garteshore, K. Pidgin, D. Sterrett, S. Tomlinson, U. Matthes and all site owners for assistance in completing this research; thanks to R. Callaway, J. Klironomos, J. Lundholm, K. McCann, U. Matthes, Pablo Garcia-Palacios, Fernando Maestre and several anonymous reviewers for helpful comments on this manuscript.