Spatial isolation slows down directional plant functional group assembly in restored semi-natural grasslands



  1. Ecological restoration schemes often assume that after reinstating appropriate abiotic conditions, plant communities will assemble following a single predictable pathway towards a fixed target state. This idea has recently been challenged, with increasing evidence that plant community assembly can only be considered deterministic at the plant trait level, rather than at the species level, and that the assembly outcome is largely influenced by the spatial context of the restoration site.
  2. We surveyed 147 vegetation plots across a chronosequence of 22 restored semi-natural grassland patches to quantify the effects of spatial isolation on both plant species and plant trait assembly. Trait level assembly was analysed using an emergent group approach, based on 28 functional plant traits. Additionally, we examined the effects of several dispersal-related plant traits on species recolonization capacities.
  3. Whereas total plant species richness of the restoration patches did not change through space or through time, progressing assembly was found to consist of a sequential replacement of generalist species with specialist species, which was reflected by a directional assembly at the plant trait level. Grassland isolation was found to slow down community assembly at both the species and the trait level without changing the general direction of assembly. This slowdown became less pronounced with increasing time since restoration.
  4. Spatial isolation of the restoration patches was found to act as a trait filter, independent of assembly age. We found a proportionally higher occurrence of species with light seeds and a high seed attachment potential in more isolated restoration patches, suggesting that colonization is more limited in isolated grasslands.
  5. Synthesis and applications. We demonstrate that the assembly process, at both the species and the trait level, is influenced by the position of the restoration patch in the landscape. Monitoring schemes following ecological restoration should therefore include the spatial context of the system while using both a trait-based and a species-based plant community analysis. Successful restoration requires physically interconnecting grassland fragments and probably the introduction of seeds or seedlings of species with low dispersability.


Large-scale habitat destruction and fragmentation have resulted in biodiversity loss and increased species rarity across Europe (Foley et al. 2005; Fischer & Lindenmayer 2007). Although much effort is currently put into the conservation of remnant habitats, passive protection of these remnants alone is unlikely to be sufficient in many landscapes to guarantee long-term persistence of many plant species, because these habitats have often become too small and too isolated (Brudvig 2011; Lindborg et al. 2011). For this reason, proper biodiversity conservation can only be accomplished through habitat restoration, focusing on the enlargement and defragmentation of the remaining habitats (Rey Benayas et al. 2009).

Because habitat restoration is crucial for conserving biodiversity, a consistent framework for restoration guidance is necessary. Restoration goals are most often embodied in a reference or target plant community that guides the restoration practices (Matthews & Spyreas 2010). This approach implicitly assumes that after re-establishment of appropriate abiotic conditions, a community will assemble following a single predictable pathway towards a fixed target state (Matthews & Endress 2010), according to the classical climax concept of succession (Clements 1916). Increasing evidence, however, shows that plant assembly does not only depend on local site conditions, but that it is also influenced by landscape characteristics and historical processes (Young, Petersen & Clary 2005; Bischoff, Warthemann & Klotz 2009; Brudvig 2011). It has been hypothesized that historical contingency, which acts through species arrival order, can lead to priority effects (Chase 2003; Fukami et al. 2005; Trowbridge 2007). These effects occur when earlier arriving species affect the establishment, growth or reproduction of later arriving species by means of competition or soil legacies. These long-lasting or even irreversible effects may result in a species composition that is different from the expected target community (Gleason 1927; Chase 2003).

Because priority effects are assumed to be independent of species identity and only to depend upon arrival sequence, identifying the factors affecting species arrival is crucial (Young, Chase & Huddleston 2001). Several approaches have been used to test whether propagule availability determines community assembly (Bischoff, Warthemann & Klotz 2009; Brudvig 2011). Seed addition experiments, for example, have demonstrated the importance of seed dispersal limitation (e.g. Turnbull, Crawley & Rees 2000; Clark et al. 2007; Hedberg & Kotowski 2010). Other studies have evaluated the effects of proximity to seed sources on species recolonization, reporting both strong distance effects (e.g. Bischoff, Warthemann & Klotz 2009; Pottier, Bédécarrats & Marrs 2009) and weak or even nonexistent distance effects (Holl & Crone 2004; Cole, Holl & Zahawi 2010; Matthews & Endress 2010). These studies have shown that landscape characteristics can have large effects on the restoration outcome, indicating that a thorough understanding of the ecological restoration process can only be accomplished through integration of the landscape context of the restoration site into the study design (Bell, Fonseca & Motten 1997; Matthews et al. 2009). Restoration sites are then conceived as isolated patches within a hostile landscape matrix, connected to mature communities through seed and propagule flow (Holl & Crone 2004; Cook et al. 2005; Young, Petersen & Clary 2005).

Whereas community assembly may have a strongly stochastic component at the species level through the action of priority effects, it has been shown that at the level of trait-based functional groups, community assembly can be considered to be more or less deterministic (Fox 1987; Fukami et al. 2005; Petermann et al. 2010). This implies that unlike species-based analyses, trait-based analyses are more likely to elucidate general assembly patterns, which may be transferable to other restoration sites, independent of site history or the taxonomic composition of the species pool (Pywell et al. 2003; Kahmen & Poschlod 2004; Pottier, Bédécarrats & Marrs 2009). Although the landscape context can be expected to have little effect on the deterministic trajectory of trait assembly, it is believed that isolation influences trait assembly in a more subtle way. It has indeed been shown that a species’ dispersal capacity is related to several species traits (Thomson et al. 2011). In this way, isolation can act as a trait filter, altering the species composition of the community (cf. Clark et al. 2007; Lindborg et al. 2011).

In the present study, we investigated the restoration process in fragmented semi-natural calcareous grasslands in southern Belgium. These extremely species-rich grasslands were once common in Europe and were maintained by regular grazing and cutting (WallisDeVries, Poschlod & Willems 2002; Pärtel, Bruun & Sammul 2005). Land-use changes during the last century have, however, led to a severe reduction in their extent (Poschlod & WallisDeVries 2002; Adriaens, Honnay & Hermy 2006). To prevent total loss of these species-rich communities, restoration of abandoned grasslands has become common practice (e.g. Poschlod et al. 1998; Butaye, Adriaens & Honnay 2005).

Previous research on these calcareous grasslands demonstrated a clear difference in the community assembly at the species level, compared to the trait level. While assembly was found to be deterministic at the trait level, it seemed to be unpredictable at the species level, with no decrease in compositional dissimilarity between different restoration patches with progressing assembly (Helsen, Hermy & Honnay 2012). Here, we wanted to study the assembly process more in depth by including the spatial context of the restoration sites and by providing practical restoration guidelines. This way we try to explore a partly neglected domain of restoration ecology by combining the plant trait response to ecological restoration with an analysis of the grassland community assembly process at the landscape scale. We used plant species abundance data from 147 plots across 22 semi-natural calcareous grasslands, restored over a 12-year time span, to answer the following questions:

  1. Does the landscape position of the restoration sites (space) and time since restoration (time) interact in mediating community assembly at the species level, making the assembly process contingent on the position of the restoration patch in the landscape?
  2. Is the previously demonstrated deterministic community assembly at the trait level influenced by the landscape position of the restoration sites?
  3. Does grassland isolation act as a trait filter, independent of restoration age, with respect to dispersal traits?

Materials and methods

Study area

The study was performed in the calcareous grassland area of the Viroin valley, in southern Belgium (see Adriaens, Honnay & Hermy (2006) for a detailed description of the study area). Restoration practices consisted of the removal of Pinus sylvestris plantations and Buxus sempervirens encroachment through cutting. After that, spontaneous colonization of the bare soil was allowed. Soil characteristics were not directly altered, nor were plant species introduced, resulting in very similar abiotic starting conditions across study sites (André & Vandendorpel 2004; Piqueray et al. 2011). After initial restoration, grassland management consisted of annual grazing by migratory sheep flocks, identical to management of the mature grasslands. Grasslands were available over a range of restoration age classes, with some grasslands consisting of a mosaic of patches of different restoration age. Restoration was carried out in two phases, first in 1995, when a small number of grassland patches were restored, and second, during a period ranging from 2001 to 2007, when larger areas were restored. In both periods, the restoration practices were identical, with newly formed grasslands adjacent to mature calcareous grasslands (Fig. 1).

Figure 1.

Study area in the Viroin valley, visualizing the sampled restored grasslands and the adjacent mature grasslands.

Species composition survey

The restored grasslands were surveyed using two 2 × 2 m plots for every restored hectare, randomly spaced over the grassland. In total, 147 plots were established in 46 restoration patches (of different restoration age, ranging from 3 to 15 years), in 22 grasslands (Fig. 1, see Table S1 in Supporting Information). Species occurrence and abundance (% cover) of all plants (tracheophytes) were recorded in the plots during spring and early summer of 2010. Abundance data were obtained by dividing the 2 × 2 m plots into four smaller 1 × 1 m subplots. Abundance was then estimated for each subplot using fixed abundance groups (1-2-5-10-15-20-25-30-…% cover). Total plot abundance of a species was calculated as the mean abundance across the four subplots.

Data analysis

Spatial and environmental variables

Four different spatial isolation metrics were calculated for each plot, based on existing vegetation maps of the study area, using qgis 1.5.0 (Quantum GIS Development Team 2010). Closest edge distance was defined as the Euclidean distance between the plot and the closest edge of the nearest mature calcareous grassland patch. The closest centroid distance was defined as the Euclidean distance between the plot and the centroid of the nearest mature calcareous grassland patch. A buffer isolation measure was defined as the total area of mature calcareous grassland present within a 1500 m radius around the plot. Finally, the Hanski isolation measure was defined as math formula (Hanski 1999), where Aj is the area of the jth mature fragment located at a distance dj from the plot, for = 1 to k, with k equal to the number of patches within a buffer of 1500 m.

To rule out the effects of environmental variation and isolation between grasslands that may covary with restoration age or isolation, we surveyed the following abiotic variables in each 2 × 2 m plot: cover (%) of bare rock and open soil, soil depth and plot inclination. These variables were averaged for each grassland. Using QGIS, the distance of each grassland to the most south-western corner of the study area was calculated, as well as its elevation and isolation (as defined previously). Except for the cover of bare soil, none of these variables were significantly correlated with grassland patch age (Table S5, Supporting Information), and they are not discussed any further. Soil depth and plot inclination were found to be significantly correlated with grassland isolation and were included in further analyses (Table S5, Supporting Information).

Emergent group delineation

Twenty-eight plant traits were selected for emergent group delineation (Table S2, Supporting Information). Traits were chosen based upon their relevance for community assembly, including the processes of dispersal, establishment and persistence (cf. Weiher et al. 1999). Trait values were retrieved from different sources (Fitter & Peat 1994; Thompson, Bakker & Bekker 1997; Bekker et al. 1998; Lambinon et al. 1998; Klotz, Kühn & Durka 2002; Poschlod et al. 2003; Kleyer et al. 2008). Two measures of seed attachment potential, one for cattle fur and one for sheep wool, were calculated according to Römermann, Tackenberg & Poschlod (2005), using information on seed morphology following Cappers, Bekker & Jans (2006). In total, 92% of all trait values were available for all species. Emergent groups were delineated following the approach of Verheyen et al. (2003), using ClustanGraphics 8 (Wishart 2006). First, the similarities between species were calculated based upon the trait values. We used Gower's similarity coefficient, because it can cope with both missing data and mixed data types (binary, ordinal, nominal and ratio) (Gower 1971). The resulting similarity matrix was used for minimum variance clustering of the species into emergent groups (Ward's method, Ward 1963). The optimal number of clusters was determined using the tree validation procedure available in ClustanGraphics 8 (Wishart 2006).

Community composition through ordination

To analyse community composition, two nonmetric multidimensional scaling ordinations (NMDS) of Bray–Curtis dissimilarity matrices were performed: one on the arcsine square root transformed plots × species matrix and one on the arcsine square root transformed plots × emergent groups matrix, using pc-ord 5.33 (McCune & Mefford 1999). To avoid convergence on a suboptimal solution, we repeated the iterative NMDS algorithm 250 times. We selected a three-dimension model based on stress reduction with a mean stress value of 17·2 and 16·6, respectively, for the species and plot ordination (McCune & Mefford 1999). The plot × emergent groups matrix was created with emergent group abundance equal to the summed abundance of all species present in that plot, belonging to the emergent group. Correlations between the plot scores on the three NMDS axes and restoration age, spatial isolation and their interaction were analysed using a linear mixed model (REML). Soil depth and plot inclination were added to the model as covariates. Because plots were located within the 22 grasslands, data were not independent. Therefore, grassland identity was included as a random factor, taking into account the spatial clustering of data within 22 independent groups. Semi-partial math formula coefficients were calculated for each covariate using the method of Edwards et al. (2008).

Diversity metrics

Species richness (S), Simpson diversity (D) and evenness (ED) were calculated for each plot, including all species, generalist species only and specialist species only. Specialist species were defined as species confined to calcareous grasslands in Belgium (Lambinon et al. 1998; Van Landuyt et al. 2006) (Table S4, Supporting Information). Species richness was also calculated for all derived emergent groups separately. Linear mixed models analogous to those performed on the NMDS plot scores were performed on all diversity indices. D and ED were logarithmically transformed to obtain a normal distribution. Species richness of specialist species, generalist species and the different emergent groups was divided by the total species richness of each plot to obtain the proportional richness, before analysis with REML.

Trait response analysis

For all species, occurring in 10–90% of the plots (108 of 247 species), the effect of restoration age and grassland isolation on their presence/absence was examined using full-factorial multivariate logistic regression models. The resulting β-coefficients of the logistic regressions for restoration age and patch isolation can be considered as a measure of the importance of these variables in driving species distribution (Dupré & Ehrlén 2002). To elucidate the mediating role of plant dispersal capacity, we evaluated the relationship between these β-coefficients and eight plant traits, using Spearman rank correlations. Plant traits used in these analyses were chosen for their relevance to the dispersal process: plant height, seed length, seed shape, seed longevity, seed mass, seed number and the earlier calculated metrics of attachment potential, one for cattle hair and one for sheep wool. All statistical analyses were performed in sas 9.1 (SAS Institute Inc. 2004).


The plot scores on the first NMDS axis, for both the species composition and emergent group composition, were significantly correlated with both restoration age and grassland isolation, with plots in the oldest and least isolated communities clustering at low values of the first ordination axis for the species ordination and with the opposite pattern apparent in the emergent group ordination graph (Table 1, Figs S1 and S2, Supporting Information). The significant interaction terms, however, indicate that differences in plot scores caused by isolation are becoming smaller with increasing grassland age (Table 1). Plot scores of the second NMDS axis were also significantly correlated with isolation for the emergent group ordination (Table 1). Because all isolation metrics were found to be intercorrelated (Table S5), only the results for the distance to the closest edge are reported. We found no effects of restoration age or grassland isolation on the diversity indices that include all species. The proportional specialist species richness, however, increased with time since restoration and decreasing isolation, indicating independent effects of restoration age and spatial isolation. The opposite was true for the proportional generalist species richness (Table 1, Figs 2 and 3). Note that the proportional generalist species richness is the one complement of the proportional specialist species richness, explaining the equal test results (Table 1). Specialist species evenness showed the opposite trend regarding the time since restoration effect and decreased with progressing assembly, indicating that an increase in specialist species number paralleled an increase in the variation of their abundance (Table 1). The opposite was true for generalist species, where a decrease in species number paralleled an increase in their evenness. However, both specialist and generalist evenness were found to be unaffected by isolation. The opposite trends for species richness vs. evenness are likely the cause of the absence of significant trends in Simpson diversity (Table 1).

Table 1. Parameter estimates of the model relating NMDS axes and diversity indices to restoration age and isolation using REML
 AgeIsolationAge × IsolationSoil depthInclination
β F math formula β F math formula β F math formula β F math formula β F math formula
  1. Beta-coefficient, test statistic and semi-partial math formula given for restoration age, isolation and the interaction term, soil depth and inclination (n = 147). Models with a nonsignificant interaction term were rerun using the main effects model. S = proportional species richness; D = logarithmic transformation of Simpson diversity; ED = logarithmic transformation of evenness.

  2. a

    See Table 2 for names and contents.

  3. Significance: *0·05 ≥ P-value > 0·01; **0·01 ≥ P-value > 0·001; ***0·001 ≥ P-value.

NMDS axis 1 (species)2·38***13·10·098−30·93*5·50·0440·015*5·50·044−0·68**7·90·0622·18*4·40·035
NMDS axis 2 (species)0·021<0·01 −0·0171·0    −0·86**8·80·0683·7**10·50·079
NMDS axis 3 (species)2·08**10·00·0760·0150·8    −0·260·7 −0·043<0·01 
NMDS axis 1 (Em. groups)2·01**6·90·054−44·34**8·50·0660·022**8·50·066−0·482·9 −0·70·4 
NMDS axis 2 (Em. groups)−1·061·2 41·19*4·50·036−0·021*4·60·0370·421·3 −1·611·4 
NMDS axis 3 (Em. groups)0·410·4 0·0110·4    0·695·1 0·490·2 
S 0·0740·08 0·00230·1    −0·0220·03 −0·022**8·60·066
S spec.0·022***23·80·16−0·00034**7·50·059   −0·0063**8·80·0680·020*6·00·047
S gen.−0·022***22·60·160·00034**7·90·061   0·0053*6·20·049−0·023**7·50·058
D −0·00900·1 0·000190·09    −0·00280·05 −0·11*6·40·050
D spec.0·00430·3 −0·000120·4    −0·00673·1 −0·028*4·40·035
D gen.−0·00240·09 0·000312·0    0·000850·04 −0·0130·8 
E D −0·00381·0 −0·000040·1    0·000200·01 −0·00470·5 
ED spec.−0·019**7·50·0590·0000430·05    0·00210·4 −0·0151·6 
ED gen.0·018*5·10·041−0·000090·2    −0·00260·5 0·042**9·40·072
S Em. group 1a−0·0056*4·50·0360·082**6·90·055−0·00004**6·90·0550·000260·08 0·0072*6·80·053
S Em. group 2a−0·0086*4·20·0340·14*6·70·053−0·00007*6·70·0530·000500·1 0·00621·3 
S Em. group 3a−0·000990·9 0·00000780·2    0·000200·2 −0·00130·5 
S Em. group 4a0·020***11·60·088−0·20**7·40·0580·00010**7·40·058−0·0038*4·10·033−0·018*6·10·048
S Em. group 5a−0·012***15·00·110·00013*4·70·038   −0·000120·01 −0·0103·4 
S Em. group 6a0·0021*4·10·033−0·00006*4·40·035   −0·000831·4 0·00170·4 
S Em. group 7a0·0062*6·30·0500·0000400·7    0·0028*6·30·0490·011*5·90·046
Figure 2.

Change in proportional species richness for specialist species through time after restoration, for grasslands with low isolation (closest edge < 75 m) (full circle, continuous line) and grasslands with high isolation (closest edge > 100 m) (open circle, dotted line). Overall mean and 95% confidence intervals are presented for each restoration age class.

Figure 3.

Correlation between the proportional species richness and fragment isolation for generalists (full circle, continuous line) and specialists (open circle, dotted line). Data points are the mean richness values for each separate grassland patch.

Seven emergent groups were obtained after inspection of the different cutting levels (Table 2, Tables S3 and S4, Supporting Information). Group names were based on the groups’ trait composition: megaphanerophytes, forest/shrub species, orchids, small grassland herbs, large herbs and grasses, sedges and shallow soil specialists and annuals. Species richness response to restoration age and isolation differed between these seven emergent groups, with significant effects of both restoration age and isolation for 5 of the 7 emergent groups, with a decrease in species richness of megaphanerophytes, forest/shrub species and large herbs and grasses and an increase in species richness of small grassland herbs and sedges and shallow soil specialists with progressing assembly and decreasing isolation (Table 1, Fig. 4). The significant interaction terms, however, indicate that differences in species richness of megaphanerophytes, forest/shrub species and small grassland herbs caused by isolation are becoming smaller with increasing grassland age (Table 1). Annuals were found to increase with progressing assembly but were unaffected by isolation. Note that the significant isolation effects in Table 1 occurred while controlling for both soil depth and plot inclination, indicating their independence of these abiotic variables. Analysing the beta-coefficients of the logistic regression equations, we found that large plants with long seeds are more likely to occur in old grasslands (Table 3) and that large plants with large and heavy seeds and a low seed attachment potential were less likely to colonize more isolated restoration patches (Table 3). The filtering caused by isolation was again found to decrease with increasing time since restoration (Table 3).

Table 2. Overview of the derived emergent groups. Emergent group name, typical plant traits and number of species are given
Emergent groupGroup nameCharacteristicsNumber of species
1MegaphanerophytesLong lived, early flowering, wind pollinated, large seeds, transient seed bank, allogamous, anemo- and dysochores. Species of nutrient-rich soils14
2Forest/shrub speciesLong lived, shade tolerant, insect pollinated, transient seed bank, mixed mating system, few and heavy seeds, dysochores, large leaves. Species of nutrient-rich soils40
3OrchidsMany, small seeds, mycorrhizal dependent13
4Small grassland herbsAllogamous, shade intolerant, small herbs, autochores and zoochores, nitrogen fixators, semi-rosette specialists68
5Large herbs and grassesSemi-rosette species, late flowering, large seeds, large leaves, hemero- and zoochores, competitives. Species of nutrient-rich soils53
6Sedges and shallow soil specialistsMixed mating system, long seed bank longevity, small and light seeds, auto- and anemochores, mycorrhizal independent28
7AnnualsEarly flowering, autogamous, short-lived, small seeds and plants, zoochores, ruderals31
Table 3. Spearman rank correlations between plant traits and logistic regression beta-coefficients for restoration age and isolation. Spearman R given for beta-coefficient for restoration age, isolation and the interaction term (n = 108). Seed number was logarithmically transformed
 Spearman R for β ageSpearman R for β isolationSpearman R for β age × isolation
  1. Significance: *0·05 ≥ P-value > 0·01; **0·01 ≥ P-value > 0·001; ***0·001 ≥ P-value.

Plant height0·24*0·21*−0·20*
Seed length0·20*0·27**−0·29**
Seed shape0·0750·0019−0·020
Seed longevity−0·042−0·120·11
Seed mass0·110·24*−0·26**
Seed number0·0700·0170·0021
Attachment potential sheep−0·0071−0·18*0·17
Attachment potential cattle−0·15−0·21*0·21*
Figure 4.

Correlation between proportional species richness and fragment isolation for five Emergent groups: Megaphanerophytes (full circle), Forest/shrub species (open circle), Small grassland herbs (grey circle), Large herbs & grasses (full triangle) and Sedges & shallow soil specialists (open triangle). Data points are the mean richness values for each separate grassland patch.


General assembly patterns

Whereas total species richness of the restoration patches did not change through time, we found directional community assembly when focusing on all other measures of community composition. These results are in accordance with previous findings in this study area, where community assembly was found to be deterministic at the trait level, but unpredictable at the species level (Helsen, Hermy & Honnay 2012). More importantly, we could identify independent effects of spatial isolation on the assembly process, with increasing isolation delaying the community assembly process (Fig. 2). This is largely in accordance with previous studies (Cook et al. 2005; Bischoff, Warthemann & Klotz 2009; Matthews & Endress 2010). It cannot be excluded that these assembly patterns through time are partly driven by changes in abiotic soil conditions (Piqueray et al. 2011).

The 15-year-old grassland plots have high leverage on our results, because a restoration time gap exists between 9 and 15 years. For this reason, the analyses were rerun after exclusion of the oldest plots. Although significance levels decreased, almost all significant patterns and conclusions remained similar, with the exception of the significant decline of generalist species through assembly.

Species level assembly

Overall species richness and diversity were found to be unaffected by time since restoration and isolation. Species composition was found to change with progressing assembly, as seen in the NMDS ordination scores. More specifically, the change in species composition involved the replacement of generalist species by specialist species. This change was counteracted by increasing isolation, indicating a slowdown of species assembly caused by spatial isolation.

The studied calcareous grasslands are known to lack a persistent seed bank (Bossuyt, Butaye & Honnay 2006), which was further confirmed in this study by the absence of a significant effect of the seed longevity index on a species’ response to grassland patch isolation. This implies that species colonization is almost fully dependent upon seed dispersal, explaining the importance of spatial isolation. This results in the longer persistence of widely available generalist species, possibly extending immigration credits (defined by Jackson & Sax (2010) as the number of species committed to eventual immigration, following a forcing event such as restoration through tree removal), because specialist species were found to have low colonization potential (Pywell et al. 2003). Although other studies indicate that isolation effects on assembly rates can become more pronounced with progressing assembly, we observed a decrease in the isolation effects with increasing time since restoration (Cook et al. 2005; Fukami et al. 2005; Trowbridge 2007). This could indicate that in this system strong priority effects will likely not occur.

Trait level assembly

At the emergent group level, progressing community assembly involved the gradual replacement of woody species and large competitive herbs and grasses with small stress-tolerant herbs, shallow soil specialists and annuals. The absence of a significant pattern for orchids is not surprising, because these species are expected to arrive late in the community assembly process (Gijbels, Adriaens & Honnay 2012). The emergent group assembly was also counteracted by isolation, with increasing isolation leading to an increase in woody species and large competitive herbs and grasses and a decrease in small stress-tolerant herbs and shallow soil specialists. The absence of an isolation effect for annual species is not so surprising, because these species are known produce a large number of highly dispersive, dormant seeds (Grime 1977).

The general direction of the emergent group assembly, however, remained unaffected by isolation. This implies that although dispersal limitation clearly affects assembly at the trait level, it does not override the functionally predictable assembly mechanisms driven by the available niches (Petermann et al. 2010). This indicates that it is possible to predict the community composition of restored calcareous grasslands at a certain point in time in terms of trait composition, but only when the position of the patch with respect to potential source patches is known.

It has been suggested that isolation can act as a dispersal trait filter that acts independently from restoration age (cf. Clark et al. 2007; Lindborg et al. 2011). We indeed found a higher incidence of small species with light seeds and a high seed attachment potential to both cattle hairs and sheep wool in highly isolated grasslands. These seed traits promote dispersal, indicating that colonization is more critical in isolated grasslands. These results are in accordance with previous studies, where low seed mass was found to have a significant effect on dispersal distance, irrespective of plant height (Thomson et al. 2011). However, unlike previous research, we did not find a positive effect of plant height on dispersal distance, probably because our analyses are restricted to grassland species, whereas plant height effects are mainly expected for tree and shrub species (Thomson et al. 2011). The significant effect of attachment potential is not surprising, because these grasslands are managed by grazing through migrating sheep flocks, which are known to function as mobile seed vectors (Poschlod et al. 1998; Adriaens, Honnay & Hermy 2007).

Implications for restoration

In accordance with our previous findings, we observed a more predictable directional community assembly at the trait level than at the species level (Helsen, Hermy & Honnay 2012). This indicates the importance of combining both approaches to clearly infer the restoration status of semi-natural grasslands (Pywell et al. 2003; Pottier, Bédécarrats & Marrs 2009; Woodcock, McDonald & Pywell 2011).

More importantly, we showed that spatial isolation slows down assembly at both the species and the trait level. At the species level, this slowdown can lead to a longer persistence of generalist species, possibly leading to strong priority effects, further altering the species assembly trajectory. At the trait level, however, this slowdown does not seem to change the predictable trajectory, with the exception of species’ dispersal traits, with isolated restoration patches biased towards species with a higher dispersal capacity. We did, however, observe that this slowdown of assembly caused by isolation will decrease with time for certain emergent groups. Nonetheless, we still believe that these findings suggest that grassland restoration will benefit from nature conservation actions that reduce dispersal limitation. This can partly be accomplished through increased cattle or sheep migration between restoration patches (Adriaens, Honnay & Hermy 2007; Hedberg & Kotowski 2010). Although seed exchange by migrating grazers likely facilitated seed dispersal in our study area, our results suggest that this is insufficient to compensate for the isolation effects on the species composition. Therefore, restoration would benefit from physically interconnecting grassland fragments or from introducing seeds or seedlings, especially of species with low dispersability (see also Brudvig 2011). If grassland connectivity remains low, successful restoration will be delayed and can result in a restored plant community that is different from the target community. This end state can be interpreted as an alternative stable state and may turn out to be very persistent against further restoration practices (cf. Suding, Gross & Houseman 2004). The incorporation of spatial configuration into restoration schemes is therefore of high importance, because it will help to predict the outcome of ecological restoration efforts.

We believe that further research should be conducted to test the validity of our results in related systems. These studies could use permanent plots surveyed over time to correct for possible undesirable effects of our chronosequence design. Moreover, measuring different species trait values in situ and allowing for intraspecies trait variation would further increase the accuracy of future studies.


This paper was written when K.H. held a grant from the Flemish Fund for Scientific Research (FWO). Thanks goes out to Kurt Hofmans, Louis-Marie Delescaille, Léon Woué, Dr. Dries Adriaens and Stijn Cornelis.