Abstract
Are there trends of increasing/decreasing dispersion of single, categorical traits related to early/late-successional species between stages of community development? If yes, are these trends dependent on species pool extension and habitat scale? Is there a consistent reduction in single trait convergence or divergence in any seral stage when scaling down from ecological to local species pool?
Presence of all vascular species rooted within plots of 5 × 5 m was recorded in assemblages of exposed mining spoils (EMS) and heathlands (HTL), which form a chronosequence on two abandoned ore tailing heaps located close to each other in the south-eastern Carpathians (Romania). Fifteen nominal, trait attributes of plant species co-occurring in the two seral assemblages were collected from available databases and subsequently classified as either successionally ‘pioneer’ or ‘mature’. The strength of single trait convergence or divergence was estimated by comparison with null plant assemblages at patch type (meta-community) level by reference to the ecological or local species pool, and at community level.
At patch type level, all pioneer and mature trait attributes (apart from short life span), with significant variation between the two seral stages, increased and, respectively, decreased in dispersion irrespective of species pool extension. However, these trends were more conspicuous when using the ecological species pool, very likely due to relaxation in abiotic filtering and dispersal limitation. At community level, no consistent trends were observed between EMS and HTL assemblages, probably because most trait attributes were sorted by microenvironmental filters displaying high variation, like topography or habitat patch geometry. In both seral stages, there was a general weakening of trait convergence or divergence at patch type level when scaling down from the ecological to the local species pool, which was due to niche space contraction. At community level, there was a trend of rise in dispersion of pioneer attributes along the observed chronosequence, presumably imputable to increasing competition for light and underground water, but an opposite trend of dispersion drop in mature attributes was not so evident. Based on these findings, we proposed two rules of thumb concerning the expected changes in dispersion of trait attributes at patch level along successions and between levels of species pool extension. In conclusion, trends in the successional dynamics of pioneer and mature trait dispersion are clearly detectable at meta-community level, especially by reference to the ecological species pool. Habitat scale and species pool extension are key factors to consider and report when estimating the magnitude of single trait dispersion.
INTRODUCTION
Ecological assembly rules aim to explain mechanistically the non-random distribution of species traits (restrictions on community structure) due to ecological filters: dispersal limitations, abiotic environment and biotic interactions (Belyea and Lancaster 1999; Götzenberger et al. 2012). Nevertheless, the relative importance of these deterministic forces is thought to vary among different stages of community development (Cottenie 2005). In the initial stages of primary succession, which starts on fully mineral substrate, abiotic factors are very strong filters whereas biotic interactions are usually of much lesser importance, but this relationship is generally reversed in the latest stages when dense canopies develop (Callaway and Walker 1997; Weiher and Keddy 1999).
The ecological filters presumably act on traits rather than on species (Keddy 1992; Keddy and Weiher 1999). Employing biological traits instead of taxonomic identities offers the possibility for direct comparisons between different seral stages that could reveal patterns of organisation in response to environmental change that are otherwise difficult to detect (Aubin et al. 2009). However, the intimate link between plant traits, environment and species interactions must be always considered in order to get insights into the operation of non-neutral assembly rules (de Bello et al. 2009; Weiher et al. 1998). Environmental filtering is likely to reduce functional trait diversity by selecting species with similar ecological characteristics (trait convergence or under-dispersion) that are able to tolerate the local environment (Díaz et al. 1998; Grime 2006; Keddy 1992; Weiher and Keddy 1995). Abiotic filtering, along with the associated trait convergence, is generally expected to be more intense in the early stages of primary successions when strong environmental constraints to plant establishment or growth are present (Weiher and Keddy 1995). On the other hand, species interactions in the initial stages of primary succession are generally of much lesser importance than in the later stages (Callaway and Walker 1997), but nevertheless can produce different patterns of trait distribution.
Negative interactions (e.g., interspecific competition), which are expected to become more prominent at later successional stages, usually select for species with different ecological characteristics (trait divergence or over-dispersion), as niche differentiation prevents coexisting species from being too similar, i.e. limiting similarity (Chesson 2000; Grime 2006; Stubbs and Wilson 2004; Weiher and Keddy 1995). Nevertheless, under particular site conditions, negative interactions can lead to trait convergence. Thus, in low fertility habitats, it is possible that size-symmetric competition for nutrients allow species with similar resource-use strategy to co-exist (Grime 2006; Kunstler et al. 2012; Mason et al. 2012; Mayfield and Levine 2010).
Positive interactions, such as facilitation, are expected to partially drive community assembly at early stages, and their importance should increase with abiotic stress (Callaway and Walker 1997; Huston and Smith 1987; Michalet et al. 2006). Since many positive interactions involve functionally distinct species (Callaway 2007), facilitation leads inevitably to trait divergence in stressful environments.
Despite the soundness of these hypotheses, empirical evidence for trait assembly processes along primary succession gradients is rare (but see Kuiters et al. 2009) and the results are inconclusive (Schleicher et al. 2011). While some of them pointed out no significant functional filtering (e.g. Schamp et al. 2008; Thompson et al. 2009), others confirmed that either functional divergence (e.g. Stubbs and Wilson 2004) or convergence occurred (e.g. Franzén 2004). Several recent studies have shown that patterns of functional trait dispersion depend on site conditions, successional stage, the traits considered, the scale of observation and how the species pool has been defined (Bernard-Verdier et al. 2012; de Bello et al. 2012; de Bello et al. 2013; Pärtel et al. 2011; Schamp et al. 2011; Spasojevic and Suding 2012). For instance, the effect of environmental filtering translated in terms of trait convergence usually decreases from regional scale (with rich, functionally diverse species pool) to local scale (with poor, functionally more redundant species pool). In addition, the landscape context or unknown stochastic factors can mask or blur the non-random trait patterns (Helmus et al. 2007; Prach et al. 2014).
Within this theoretical framework, we aimed at detecting patterns in plant trait distribution that should represent signatures of different assembly processes (dispersal limitation, habitat filtering and plant interactions) operating in two early-mid stages of primary succession. For this purpose, we inventoried two ore-mining spoil heaps of different ages but displaying similar and relatively homogeneous substrate. In particular, we estimated the strength and significance of patterns in plant trait distribution by comparing the observed communities with null plant assemblages, and attempted to infer trends between the two seral stages by taking into account the species pool extension and the habitat scale.
In this article, we questioned the following hypotheses: (i) Are there trends of decreasing/increasing frequency of traits related to early/late-successional species between stages of community development due to weakening in abiotic filtering and dispersal limitation? (ii) Do these trends weaken from higher to lower levels of habitat scale (meta-community versus community) and species pool extension (landscape versus local)? (iii) Is there a consistent reduction in trait convergence and divergence attributable to niche space contraction in any seral stage when scaling down from ecological to local species pool? (iv) Does the dispersion of early and late-successional traits rise and respectively, decline from early to mid-successional stage due supposedly to increasing competition? Moreover, we attempted to detect some of the environmental factors that accounted for the filtering of single traits.
MATERIALS AND METHODS
Study area
Roșia Montană (46°18′0″N, 23°08′0″E) is situated in the Metalliferous Mountains (south-eastern Carpathians, Romania), within the gold-mining district known as the Golden Quadrilateral. Despite the mine waste dumps present in the area represent a harsh environment for colonising plants due to high content of metals and lack of humus, they were particularly suitable for our study because the ore tailing heaps were abandoned at different times. We selected for investigation two of them, which were 35 (Iuliana) and, respectively, 60-year-old (Orlea) at the time of field work (2009). Both tailing dumps were covered by a mosaic of vegetation types including pioneer assemblages of exposed mine spoils, grassland, heathland (with Calluna vulgaris, Vaccinium myrtillus and V. vitis-idaea) and open woodland (mainly Betula pendula, Populus tremula and Pinus sylvestris), but in contrasting proportions. This pattern points out that overall the underlying mechanism of such succession has roughly followed the passive tolerance model, which states that colonisation can start with any species from the available pool through possession of contrasting life histories (Pickett et al. 1987). The different age but common substrate and close proximity (only 0.6 km apart) of the two post-mining dumps allowed a space-for-time reconstruction of primary succession advancement under the reasonable assumption of a common regional species pool. For the purpose of this study, we chose to analyse a chronosequence of two stages represented by patches of exposed mine spoils (EMS) on Iuliana Heap (about 18% of the area) and patches of heathland (HTL) on Orlea Heap (about 25% of the area).
Floristic data collection
A stratified random sampling was performed within an area of 500 × 500 m covering each mine tailing heap. An equal number of plots (50) was assigned within patches of EMS (on younger heap) and HTL (on older heap), but the number of plots per patch was proportional to its size. The presence of each vascular plant species rooted within a plot was recorded in 5 × 5 m quadrats. Subsequently, all vascular plant species occurring throughout the whole area of 0.25 ha on each waste heap were recorded through a systematic inventory performed in grid cells of 50 × 50 m. We used the list of all vascular species inventoried (192 and 161 taxa on Iuliana and, respectively, Orlea dump) to build the so-called ecological species pool (Belyea and Lancaster 1999), whose members are limited exclusively to those species that overcame both dispersal and habitat constraints. The two ecological species pools distinguished at the two sites shared 51% of the total number of species recorded.
Trait data collection
We employed only relatively ‘stable’ traits whose intraspecific variation is low compared with their interspecific variation (Garnier et al. 2007). Moreover, the biological traits are among the characteristics best correlated with species success in succession (Prach and Pyšek 1999). As a consequence, we selected a series of categorical traits (from those available for the majority of species) that we retained relevant in terms of their association with the early and late-successional species: growth-form, life span, mode of reproduction, vegetative propagation organs, dispersal modes, leaf disposal, nutrition adaptation and ploidy type. In most cases, we did not use all possible attributes (states) of single traits, so that to reduce redundancy and to gain power of discrimination. For this reason, some species did not fit in any of the categories retained for one trait. A total of 15 nominal, trait attributes were distinguished, of which eight were considered to be associated with early successional species (‘pioneer attributes’) and seven with late successional species (‘mature attributes’), as suggested by current theories and empirical evidence (Table 1). Most of these data were collected online from trait databases, i.e. Ecological Flora (Fitter and Peat 1994), BiolFlor (Klotz et al. 2002) and BIOPOP (Poschlod et al. 2003). Subsequently, the gaps revealed in trait data were filled with information from literature, apart from two rare, sub-endemic species for which no data on ploidy type were found. To avoid the removal of the two species from analyses that do not allow for missing data, we assigned to each of them the category corresponding to the conservative scenario with respect to current theories (te Beest et al. 2012), i.e. polyploid to the species occurring exclusively in HTL and not polyploid to the one recorded only in EMS patches.
: empirical predictions of dispersion patterns in single, pioneer and mature trait attributes within early (EMS) and mid (HTL) seral assemblages
| Seral age | Plant trait attribute | Attribute abbreviation | Pattern expected | |
|---|---|---|---|---|
| EMS | HTL | |||
| Pioneer (early successional) | Anemochory | ANM | + | 0 |
| Polyploidy | PPD | + | − | |
| Basal leaf rosette | RST | + | − | |
| Runners (adventitious roots and stolons) | Run | + | + | |
| Reproduction mostly via seeds | Seed | + | + | |
| Short-lived (annuals and biennials) | S_L | + | − | |
| Grass | Grass | + | − | |
| Nitrogen-fixing | N_fix | + | − | |
| Mature (late successional) | Zoochory | Zoo | − | + |
| Root–shoot regeneration mode | R_S | − | + | |
| Underground storage (rhizome, tuber or bulb) | USO | − | + | |
| Reproduction mostly vegetative | Veg | − | + | |
| Forb | Forb | 0 | + | |
| Shrub | Shrub | − | + | |
| Tree | Tree | − | − | |
| Seral age | Plant trait attribute | Attribute abbreviation | Pattern expected | |
|---|---|---|---|---|
| EMS | HTL | |||
| Pioneer (early successional) | Anemochory | ANM | + | 0 |
| Polyploidy | PPD | + | − | |
| Basal leaf rosette | RST | + | − | |
| Runners (adventitious roots and stolons) | Run | + | + | |
| Reproduction mostly via seeds | Seed | + | + | |
| Short-lived (annuals and biennials) | S_L | + | − | |
| Grass | Grass | + | − | |
| Nitrogen-fixing | N_fix | + | − | |
| Mature (late successional) | Zoochory | Zoo | − | + |
| Root–shoot regeneration mode | R_S | − | + | |
| Underground storage (rhizome, tuber or bulb) | USO | − | + | |
| Reproduction mostly vegetative | Veg | − | + | |
| Forb | Forb | 0 | + | |
| Shrub | Shrub | − | + | |
| Tree | Tree | − | − | |
Convergence and divergence are indicated by plus and respectively, minus signs (0 = neutral).
: empirical predictions of dispersion patterns in single, pioneer and mature trait attributes within early (EMS) and mid (HTL) seral assemblages
| Seral age | Plant trait attribute | Attribute abbreviation | Pattern expected | |
|---|---|---|---|---|
| EMS | HTL | |||
| Pioneer (early successional) | Anemochory | ANM | + | 0 |
| Polyploidy | PPD | + | − | |
| Basal leaf rosette | RST | + | − | |
| Runners (adventitious roots and stolons) | Run | + | + | |
| Reproduction mostly via seeds | Seed | + | + | |
| Short-lived (annuals and biennials) | S_L | + | − | |
| Grass | Grass | + | − | |
| Nitrogen-fixing | N_fix | + | − | |
| Mature (late successional) | Zoochory | Zoo | − | + |
| Root–shoot regeneration mode | R_S | − | + | |
| Underground storage (rhizome, tuber or bulb) | USO | − | + | |
| Reproduction mostly vegetative | Veg | − | + | |
| Forb | Forb | 0 | + | |
| Shrub | Shrub | − | + | |
| Tree | Tree | − | − | |
| Seral age | Plant trait attribute | Attribute abbreviation | Pattern expected | |
|---|---|---|---|---|
| EMS | HTL | |||
| Pioneer (early successional) | Anemochory | ANM | + | 0 |
| Polyploidy | PPD | + | − | |
| Basal leaf rosette | RST | + | − | |
| Runners (adventitious roots and stolons) | Run | + | + | |
| Reproduction mostly via seeds | Seed | + | + | |
| Short-lived (annuals and biennials) | S_L | + | − | |
| Grass | Grass | + | − | |
| Nitrogen-fixing | N_fix | + | − | |
| Mature (late successional) | Zoochory | Zoo | − | + |
| Root–shoot regeneration mode | R_S | − | + | |
| Underground storage (rhizome, tuber or bulb) | USO | − | + | |
| Reproduction mostly vegetative | Veg | − | + | |
| Forb | Forb | 0 | + | |
| Shrub | Shrub | − | + | |
| Tree | Tree | − | − | |
Convergence and divergence are indicated by plus and respectively, minus signs (0 = neutral).
Environmental and patch geometry data acquiring
We digitised all vegetation patches covering the two mine spoil heaps in ArcGIS software (version 9.3.1, Environmental Systems Research Institute, Redlands, CA, USA) by using geo-referenced topographical maps, aerial photographs (at scale 1:5000) with a cell size (pixel) of 0.5 m, and additional ground information. A digital 2 m elevation model was used to determine topographic metrics for each plot.
The mean slope, aspect and terrain curvature at plot level, as well as the perimeter, area and ecotones length of all EMS and HTL patches, were obtained directly in ArcGIS using the Spatial Analyst Tool and the Hawth’s Tools extension (version 3.27, www.spatialecology.com/htools). The terrain curvature was expressed on a dimensionless, unbound scale ranging from negative values (proportional to concavity) to positive values (proportional to convexity), with zero corresponding to flat ground. Solar radiation energy per unit area was estimated in each plot by involving latitude, slope and aspect in ArcGIS using the Solar Radiation toolset.
Data transformation and analysis
All nominal trait data were transformed in binary variables, one for each attribute (state) distinguished. To avoid spurious correlations with other variables and to allow direct comparisons, the length of ecotones around each EMS and HTL patch was expressed as fractions of its total perimeter. Transformation of variables was applied before fitting linear models, so that to reduce positive skewness and/or heteroscedasticity. All environmental data were square-root transformed, except for those variables expressing proportions that underwent an angular (arcsine) transformation.
Unconstrained, distance-based redundancy analysis was employed to explore the separation of pioneer and mature trait attributes as well as to estimate the amount of redundancy in trait data. The trait attributes were scaled proportionally to eigenvalues in the species space defined on the basis of the complement of Sørensen similarity.
Convergence and divergence were estimated on single trait attributes separately (e.g. Cornwell and Ackerly 2009; Kraft et al. 2008) as the multiple assembly processes, which operate simultaneously to structure plant communities, would be obscured by a single multivariate trait index (Spasojevic and Suding 2012). Two different measures were employed to estimate trait convergence or divergence: the frequency of single trait attribute across all plots at each study site and the mean trait-dispersion between species co-occurring in the same plot. The first measure was useful for detecting filters at the interface between the species pool (either the ecological species pool or the local one) and the habitat of interest, i.e. EMS or HTL taken as a whole (meta-community). The second measure (TAUst), which is the trait-based analogue of the phylogenetic index PIst as defined by Hardy and Senterre (2007), was particularly adapted for distinguishing patterns related to filters operating between distinct plots (but ascribed to patches of the same type) or related to processes acting within communities. The statistical significance of the observed metrics in real communities was expressed in terms of their standardised effect size (SES) by reference to the means and standard deviations of simulated indices drawn from 10000 random species assemblages (Gotelli and McCabe 2002). Assuming a normal distribution of SES based on the central limit theorem, values larger than 2 and lower than −2 were roughly considered different from zero. Based on differences between observed and simulated statistics, one of the three patterns could be detectable for each trait attribute: convergent (underdispersion), divergent (overdispersion) or neutral; the latter occurs when either trait-neutral processes are operating or opposing selective forces are balanced (Schamp and Aarssen 2009). Three types of randomisation algorithms were used to generate null models of community assembly, depending on the species pool distinguished and the hypothesis being tested.
Trait dispersion analysis was performed at two scales: patch type level (by reference to the ecological or local species pool) and community level (by reference to the local species pool). At patch type level, we sought to reveal the effects of abiotic filtering, on one hand and dispersal limitation/niche differentiation, on the other hand, by comparing the observed frequencies of single trait attributes across all plots with those simulated through a particular model of community assembly: species were randomly assigned to null communities, after being selected from the ecological species pool with probabilities that matched their frequencies of occurrence within the study site, until the observed species richness in each plot was reached. A similar approach and algorithm was also employed to run a simulation by selecting species from the local species pool with probabilities that matched their frequencies of occurrence in the communities sampled. This way we could compare the effect of species pool extension on different trait patterns.
At community level, we searched for the signature of microhabitat filtering that could operate between communities from patches of the same type. For this purpose, we compared the observed mean dispersion (TAUst) values for single trait attributes with those generated by a null model that kept the community matrix unaltered but affected the trait-by-species structure, i.e. randomly assigning traits among species while maintaining the trait frequencies fixed (Stubbs and Wilson 2004). Finally, to reveal possible trait convergence or divergence due to plant–plant interactions, we used again the observed mean TAUst values for comparison with those simulated through a null model that only altered the internal structure of the community matrix (maintaining both species frequencies and richness fixed), while the trait matrix was kept unchanged. To guess what kind of interspecific interactions could be responsible for the observed patterns in single attribute dispersion, the relative number of positive and negative associations between species were estimated using a probabilistic model of species co-occurrence (Veech 2013).
Generalised linear mixed models (GLMMs) with negative binomial distribution (via log link) were employed to model single trait attribute frequency at plot level as a function of geometric and topographic metrics of the surrounding patch (fixed effects), while controlling the influence of spatial autocorrelation. The latter was handled through a residual random component with an exponential spatial covariance structure, which is based on the Euclidean distances between plot centroids. The selection of variables in GLMMs was made through a manual, stepwise procedure but starting with patch perimeter as a covariable in the model (perimeter was highly correlated with patch area but always a better predictor than the latter). The goodness-of-fit of each model was assessed through generalised chi-square divided by the degrees of freedom (DF). The inclusion of interaction terms between the variables selected in the final models did not result in additional significant effects or better fit.
All numerical analyses were performed in R (version 3.2.2, www.R-project.org) using the packages ‘vegan’ (version 2.3-1), ‘spacodiR’ (version 0.13.0115) and ‘cooccur’ (version 1.2), except for GLMMs that were run in SAS/STAT (version 9.3, SAS Institute Inc., Cary, NC, USA).
RESULTS
Correlation structure of trait attributes
The empirical differentiation between pioneer and mature traits was supported, to a certain extent, by their grouping in species ordination space (Fig. 1). Thus, in both seral stages there were positive correlations between ‘Anemochory’ and ‘Grass’, ‘Underground storage’ and ‘Forb’, ‘Polyploidy’ and ‘Runners’, ‘Root-Shoot’ and ‘Shrub’, but also negative correlations between ‘Reproduction via seeds’ and ‘Runners’, ‘Forb’ and ‘Grass’ or ‘Anemochory’ and ‘Zoochory’ (Fig. 1). Nevertheless, the set of trait attributes selected did not appear so redundant, as the proportion of variance explained by the first two RDA axes was 39% in EMS and respectively, 37% in HTL.
: unconstrained distance-based RDA ordination of trait attributes in the species’ space corresponding to EMS and HTL seral assemblages (pioneer and mature trait attributes in italics and bold, respectively). The first two axes explain 39% (EMS) and respectively, 37% (HTL) of the total variance. Abbreviations as in Table 1.
Trait filtering from the ecological species pool at patch type level
Apart from ‘Short-lived’ that displayed an opposite trend, all the pioneer attributes considered showed a significant increase in divergence or decrease in convergence (lower SES) in HTL compared with EMS (Fig. 2), which was probably due to reduction in specific niche availability. Conversely, a significant decrease in divergence or increase in convergence (higher SES) in HTL over EMS was observed in all mature traits (Fig. 2), most likely because of niche differentiation and specialisation.
change between EMS and HTL stages in SES of cumulated frequencies of pioneer (p) and mature (m) trait attributes at patch type level by reference to the ecological species pool. Abbreviations as in Table 1.
Trait filtering from the local species pool at patch type level
Of pioneer trait attributes, only ‘Runners’ and ‘Polyploidy’ showed a significant decrease in frequency from EMS to HTL stage (Fig. 3). On the other hand, three mature trait attributes (‘Veg’, ‘Shrub’ and ‘Tree’) exhibited a significant rise in frequency towards convergence and other two (‘Underground storage’ and ‘Forb’) underwent a significant decline in divergence between the two stages of community development (Fig. 3). Most of these patterns can be attributed to the higher abilities of late-successional species to reproduce vegetatively. The rest of trait attributes did not exhibit significant SES in any stage.
change between EMS and HTL stages in SES of cumulated frequencies of pioneer (p) and mature (m) trait attributes at patch type level by reference to the local species pool. Dashed lines indicate non-significant changes within the range of null effect size (−2 to +2). Abbreviations as in Table 1.
Trait patterns and environmental filters at community level
‘Anemochory’ was the only pioneer trait attribute that displayed a significant decrease in convergence along the chronosequence, whereas the others underwent non-significant changes (Fig. 4). Of mature trait attributes, only ‘Forb’ and ‘Underground storage’ showed a significant increase in convergence from EMS to HTL assemblages, whereas in contrast, ‘Tree’ declined significantly in convergence between the two seral stages (Fig. 4). The remaining mature trait attributes did not exhibit variations outside the confidence interval of null effect.
No significant overdispersed trait attributes (SES < −2) were detected in either stage at community level (Fig. 4), suggesting the absence of dispersal constraints.
In EMS assemblages, both ‘Anemochory’ and ‘Reproduction via seeds’ were negatively and positively related to the relative length of ecotones with grasslands and woods, respectively (Table 2). The length of patch perimeter had a positive effect on ‘Zoochory’ in EMS and on ‘Underground storage’ in HTL (Table 2). Finally, the frequency of forbs dropped with increasing solar radiation and terrain curvature in EMS, but rose with solar radiation in HTL (Table 2).
GLMM standardized coefficients of the (fixed) effects of patch geometry, adjacency and topography on the frequencies of single trait attributes by controlling the spatial autocorrelation (only models displaying significant effects are shown)
| Model terms and statistics trait attribute | Perimeter | RLWE | RLGE | Solar radiation | Terrain curvature | Intercept | Generalised chi2/DF |
|---|---|---|---|---|---|---|---|
| Anemochory (in EMS) | 4.97** | −3.65* | −5.06*** | 1.0 | |||
| Reproduction mostly via seeds (in EMS) | 5.90** | −5.14* | −5.95*** | 1.0 | |||
| Zoochory (in EMS) | 6.94* | −7.69*** | 1.0 | ||||
| Forb (in EMS) | −3.25* | −5.29* | −7.05*** | 1.0 | |||
| Underground storage (in HTL) | 15.92** | −11.57*** | 1.0 | ||||
| Forb (in HTL) | 10.22** | −10.81*** | 1.0 |
| Model terms and statistics trait attribute | Perimeter | RLWE | RLGE | Solar radiation | Terrain curvature | Intercept | Generalised chi2/DF |
|---|---|---|---|---|---|---|---|
| Anemochory (in EMS) | 4.97** | −3.65* | −5.06*** | 1.0 | |||
| Reproduction mostly via seeds (in EMS) | 5.90** | −5.14* | −5.95*** | 1.0 | |||
| Zoochory (in EMS) | 6.94* | −7.69*** | 1.0 | ||||
| Forb (in EMS) | −3.25* | −5.29* | −7.05*** | 1.0 | |||
| Underground storage (in HTL) | 15.92** | −11.57*** | 1.0 | ||||
| Forb (in HTL) | 10.22** | −10.81*** | 1.0 |
RLWE, relative length of woodland ecotone; RLGE, relative length of grassland ecotone.
*0.01< P < 0.05; **0.001 < P < 0.01; ***P < 0.001.
GLMM standardized coefficients of the (fixed) effects of patch geometry, adjacency and topography on the frequencies of single trait attributes by controlling the spatial autocorrelation (only models displaying significant effects are shown)
| Model terms and statistics trait attribute | Perimeter | RLWE | RLGE | Solar radiation | Terrain curvature | Intercept | Generalised chi2/DF |
|---|---|---|---|---|---|---|---|
| Anemochory (in EMS) | 4.97** | −3.65* | −5.06*** | 1.0 | |||
| Reproduction mostly via seeds (in EMS) | 5.90** | −5.14* | −5.95*** | 1.0 | |||
| Zoochory (in EMS) | 6.94* | −7.69*** | 1.0 | ||||
| Forb (in EMS) | −3.25* | −5.29* | −7.05*** | 1.0 | |||
| Underground storage (in HTL) | 15.92** | −11.57*** | 1.0 | ||||
| Forb (in HTL) | 10.22** | −10.81*** | 1.0 |
| Model terms and statistics trait attribute | Perimeter | RLWE | RLGE | Solar radiation | Terrain curvature | Intercept | Generalised chi2/DF |
|---|---|---|---|---|---|---|---|
| Anemochory (in EMS) | 4.97** | −3.65* | −5.06*** | 1.0 | |||
| Reproduction mostly via seeds (in EMS) | 5.90** | −5.14* | −5.95*** | 1.0 | |||
| Zoochory (in EMS) | 6.94* | −7.69*** | 1.0 | ||||
| Forb (in EMS) | −3.25* | −5.29* | −7.05*** | 1.0 | |||
| Underground storage (in HTL) | 15.92** | −11.57*** | 1.0 | ||||
| Forb (in HTL) | 10.22** | −10.81*** | 1.0 |
RLWE, relative length of woodland ecotone; RLGE, relative length of grassland ecotone.
*0.01< P < 0.05; **0.001 < P < 0.01; ***P < 0.001.
Trait pattern change between nested levels of species pool
In both seral stages, there was a common trend of decreasing in absolute value of the significant SES, when scaling down from ecological to local species pool (Fig. 5). That means a weakening of trait convergence or divergence when ecological filtering occurs from a subset of the species pool.
changes in SES of cumulated frequencies of pioneer (p) and mature (m) trait attributes at EMS (upper graphs) and HTL (lower graphs) patch type level that are induced by the extension of species pool (ESP and LSP denote the ecological and local species pool, respectively). Other abbreviations and line patterns as in Table 1 and Fig. 3, respectively.
Trait patterns due to plant–plant interactions
The mean dispersion of trait attributes in both stages did not have any significant, negative SES (Fig. 6), which is an indication of the lack of trait divergence.
There were two significantly underdispersed (pioneer) attributes in EMS (‘Anemochory’ and ‘Reproduction via seeds’) that turned non-significant in HTL (Fig. 6). A conspicuous drop in convergence between the two stages was also recorded for ‘Tree’ (Fig. 6). ‘Forb’ was the only mature attribute whose underdispersion increased from EMS to HTL (Fig. 6).
Whereas the proportion of positive species association was not different from random in either stage, the relative number of species pairs displaying significant, negative relationships on EMS was almost four times as high as in HTL (Fig. 7). This suggests that the higher number of underdispersed attributes revealed in EMS assemblages may be linked to the larger proportion of negative association between species.
proportion of negative and positive pairwise associations between species based on their co-occurrence at plot level in EMS and HTL assemblages. Only the number of species pairs displaying negative associations in EMS is significantly larger than expected in null communities (P < 0.0001).
DISCUSSION
Successional trends in trait dispersion at patch type level
There was no consistent response (in terms of under and over-dispersion) of either pioneer or mature attributes to ecological filters in any of the two seral stages. There are several potential reasons for such an inconsistency, among which trade-offs in plant traits, dispersal barriers, particular microsite conditions, environmental microheterogeneity of habitat patches or different stochastic factors. The only feature shared by all trait attributes was that each one showed significant convergence or divergence in at least one seral stage, which suggests that the traits selected were appropriate for tracking non-neutral processes involved in community assembly.
Concordant to the first hypothesis questioned, there was a trend of increasing and decreasing divergence in pioneer and respectively, mature attributes between the early-mid seral stages. This trend is very likely due to relaxation in abiotic filtering and dispersal limitation that operate contrasting changes depending on the group of attributes filtered (pioneer versus mature). The sole exception was represented by the short-living plants that were strikingly under-represented on the EMS. One possible, but not fully satisfactory explanation could be that the perennial herbaceous species developing above-ground runners, along with the woody species, may have overcome the short-living species that emerged from seeds. The unexpected over-representation of trees in the early stage of primary succession is probably related to the large grain size of mining spoils deposited on the younger heap. As observed on uranium-mining wastes in course of colonisation, pioneer trees (birches, especially) can germinate and establish better than herbs under conditions of coarse substrate texture (Sänger and Jetschke 2004). On top of that is the capacity of all tree species occurring on EMS (Betula pendula, Populus tremula, Salix caprea and Pinus sylvestris) to reproduce easily through wind-dispersed seeds.
A quite different picture of trait frequency patterns was disclosed when taking into account the local species pool. Significant convergence or divergence was detected in much fewer attributes, which is consistent with the second hypothesis questioned. That pattern was more evident in EMS and in the case of pioneer attributes, which is very likely the consequence of a more homogeneous source habitat (especially EMS), when considering only species dispersed from patches of the same type. Nevertheless, as predicted by the first hypothesis questioned, there was a tendency of decreasing in frequency in pioneer attributes as well as a coherent rise in frequency among mature attributes between the two seral stages.
Successional trends in trait dispersion at community level
Unexpectedly, no coherent trends in trait dispersion along succession were observed at community level by reference to local species pool. That means a much stronger effect of habitat scale reduction than that envisaged by the second hypothesis questioned, i.e. trend fading compared with trend weakening. The only trait attributes that changed their dispersion according to the first hypothesis questioned were ‘Anemochory’ (decrement), ‘Underground storage’ and ‘Forb’ (increment). The latter two were presumably related to the more favourable substrate conditions in HTL, especially in terms of organic layer development and belowground moisture retention. The unpredicted decline in ‘Tree’ convergence was probably the consequence of the reduction in regeneration niche for wind-dispersed species, given that all early colonising trees were anemochorous.
Regarding the rest (majority) of trait attributes, the trivial variation in their dispersion between the two seral stages was a direct consequence of the random patterns observed in the distribution of these traits at community level. It is possible, however, that at least some of these random patterns were solely the outcome of the noise effect of microenvironmental heterogeneity and landscape context, and not of neutral processes (Helmus et al. 2007; Prach et al. 2014).
The most striking difference between the two levels of habitat scale, when filtering traits from the local species pool, was the lack of any over-dispersed trait attribute at community level in both seral stages. That means that dispersal limitation did not operate at such (smaller) scale, in opposition to what observed at patch type level. Additional microenvironmental differences between communities (plots) were very likely responsible for the decrease in dispersion in some trait attributes: ‘Anemochory’ and ‘Tree’ (in EMS) and ‘Underground storage’ and ‘Forb’ (in HTL).
Importance of species pool extension
In accordance with the third hypothesis questioned, there was a consistent decline in trait convergence or divergence when scaling down from ecological to local species pool. This is because of the shrunken niche space associated with the (poorer, more homogeneous) local species pool as compared with that of the ecological species pool, including higher similarities in reproduction and dispersal abilities between species occurring in patches of the same type. A similar increase in convergence, when using an extended species pool as opposed to the reduced (local) species pool, was also observed by Chalmandrier et al. (2013) but the results are not comparable due to major methodological differences. The larger number of trait attributes displaying random patterns, which was observed when using the local instead of the ecological species pool, suggests that neutral processes were more important at smaller scales. However, that was more obvious in the younger spoil heap (Iuliana).
Environmental filters and traits sorted at community level
Dispersal and reproduction attributes were mostly sorted by a special category of environmental filters that were part of the landscape context. The EMS ecotones with grasslands and woodlands seem to have acted as seed traps and respectively, sources of propagules for plants that reproduce almost exclusively via wind-dispersed seeds. Similar effects of habitat adjacency in fragmented landscape but on plant species composition and richness of target patches were also observed elsewhere (Fuller and del Moral 2003; Hersperger and Forman 2003; Lawson et al. 1999). Moreover, HTL patches displaying more elongated shapes and/or more jagged edges hosted an excess number of zoochorous plants and herbs producing underground storage organs. These relationships suggest that longer boundaries increase the seed influx and vegetative lateral spread through larger ecological flows between adjacent habitats, e.g. animal movements, surface water run-off and landslides (Hansen and di Castri 1992; Prach and Pyšek 1999).
Congruent with other observations on the response of forb richness to topographic conditions (e.g., Bruun et al. 2006), EMS assemblages developed on concave terrain and/or shady slopes hosted a higher number of forbs; both topographic factors are examples of small-scale abiotic filters detectable at community level. However, under the shade of shrubs and scattered trees (in HTL), the outcome of filtering changed to the opposite, given the decline in the frequency of forbs with the decrease in solar radiation.
Contribution from plant–plant interactions
In accordance with the forth hypothesis questioned, there was a trend of dispersion rising in pioneer attributes, presumably due to increasing competition for light and underground water, from early to late successional stages. However, an opposite trend of dispersion drop in mature attributes was not so evident. The big exception was represented by trees, most of which belonging to pioneer species, that had a remarkable dispersal and establishment success on EMS. The obscurity of these trends was probably determined by the relatively large size of sampling plots, as patterns in trait dispersion due to species interactions might be plainly detected at much smaller scale, i.e. between neighbouring plants (Murrell and Law 2003; Stubbs and Wilson 2004). That is probably why we did not observe significant divergence in any trait attribute and seral stage. Another possible explanation for such inconspicuous trends might be the unexpectedly large number of negative species associations in EMS assemblages compared with HTL, although species segregation cannot be directly imputable to interspecific relationships.
At community scale, there was no evidence of increasing importance of plant species interactions from early to mid-successional stage, as suggested by the higher number of significantly under-dispersed attributes revealed in EMS compared with HTL. The negative interactions between species in both stages were probably responsible for the convergence in both growth forms ‘Forb’ and ‘Tree’, through the mechanism of size-symmetric competition (Grime 2006; Kunstler et al. 2012; Mason et al. 2012; Mayfield and Levine 2010). Their competitive advantages could be attributed to, but not necessarily limited to, the shade tolerance and underground storage organs developed in forbs, and the higher stature of trees.
It is plausible that other kind of biotic interactions could have also contributed to the observed trait patterns. Allelopathic interference between different plant species as well as symbiotic relationship between trees and mycorrhizal fungi may have played a role not at all negligible within EMS assemblages. For instance, root infection with mycorrhizae is common among pioneer trees (especially, birches) colonising mining spoil deposits (Sänger and Jetschke 2004).
Proposal of two rules of thumb
The trends revealed in trait dispersion variation along primary succession were useful for inferring two rules of thumb, in which the habitat scale and/or the extension of species pool play a key role.
1. Single, pioneer and mature trait attributes should increase and respectively, decrease in dispersion at patch type (meta-community) level along the successional gradient, irrespective of species pool extension.
Such trends are, nevertheless, more conspicuous when filtering from the ecological species pool. While the dispersion in each pioneer attribute probably increases monotonically along the whole successional gradient, the decline in dispersion in some mature attributes may flatten and eventually rise towards the late seral stages as a consequence of limiting similarity. This hypothesis is empirically based on the decrease in the relative importance of environmental filtering relative to niche differentiation, which in turn makes functionally similar species be replaced by functionally more distinct ones as succession proceeds (Purschke et al. 2013).
2. There should be a decline in magnitude of single trait convergence and divergence at patch type (meta-community) level when scaling down from ecological to local species pool in any seral stage.
The more dissimilar (in richness and ecological heterogeneity) the compared species pools are, the stronger the effect on trait dispersion is expected to be.
Both rules apply when communities belonging to the same vegetation type and meta-community are considered. This is particularly important because species randomisation in null models must be constrained within rather homogeneous environmental conditions and between non-isolated communities, in order to disentangle abiotic filtering from limiting similarity (Mayfield and Levine 2010; Weiher and Keddy 1995).
Despite these empirical findings, exceptions from the two rules of thumb stated above can occur because both convergence and divergence are still dependent, among others, on traits considered, site conditions, microenvironmental heterogeneity, landscape context and dispersal vectors and both can be altered or masked by hidden disturbance. All these inter-dependencies may be responsible for the contrasting patterns of trait convergence and divergence reported in the literature.
Limitations and concluding remarks
Our results are confined to nominal, presumably invariant traits, to one landscape context and to an incomplete, tolerance model-based successional pathway inferred from a two-stage chronosequence (late, climax stage missing in the study area). The null model approach, although powerful, cannot distinguish between pure neutral processes and null effects due to opposing forces that balance each other. Although the patterns of trait dispersion seem to be habitat scale-dependent, we could not assess rigorously the effect of scaling down from meta community (patch type) to community level, because of the different statistics and null models involved in computations. It must be also stressed that the distinction between pioneer and mature traits is merely theoretical and not all plant trait attributes can be unequivocally assigned to one of the two categories. Further studies, involving different traits, habitats and alternative methods, are needed to validate our findings.
Despite these limitations, trait dispersion analysis carried out across seral stages of primary succession can provide valuable insights into the patterns of trait assembly under the effects of abiotic and biotic filtering. In particular, the distinction between pioneer and mature trait attributes allowed for inferring trends in their dispersion between seral stages. Moreover, the use of different, nested species pools as reference in parallel analyses, performed on the same seral stage, had the merit to reveal the influence of niche space extension on the effect size of trait convergence or divergence.
ACKNOWLEDGEMENTS
This work was supported by the Romanian National Authority for Scientific Research and Innovation [PN 16 19 BIODIVERS].
Conflict of interest statement. None declared.
