SEARCH

SEARCH BY CITATION

Keywords:

  • Assembly rules;
  • Dispersal limitation;
  • Environmental filtering;
  • Functional traits;
  • Seed mass;
  • SLA

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

Question

The assembly of plants into communities is one of the central topics in plant community ecology. The objective of this study was to investigate how plant functional trait diversity and environmental factors influence community assembly in two different grassland communities, and if variation in these factors could explain the difference in species assembly between these communities.

Location

Six grazed ex-arable fields and eight semi-natural grasslands in southeast Sweden.

Methods

We estimated species abundance and measured soil attributes at each site. For each species within each site we measured specific leaf area (SLA), leaf dry matter content (LDMC) and seed mass. We analysed the data both for abundance-weighted species values and species occurrence.

Results

Trait gradient analysis indicated random distribution of species among sites, while CCA analysis indicated that both soil phosphorus and moisture were related to species assembly at a site. Correlations and fourth-corner analysis also revealed a relationship between measured species traits and soil phosphorus and moisture. There was a lower average seed mass and higher SLA of species in ex-arable fields compared to species in semi-natural grasslands.

Conclusions

Even though trait gradient analysis indicated that plant community assembly in the studied grasslands was random, other results implied that species occurrence and abundance was influenced both by environmental factors and species traits. Higher species richness in semi-natural grasslands was associated with more large-seeded species found there compared to ex-arable fields, indicating that large-seeded species establish in grasslands later than small-seeded species.


Nomenclature
Mossberg & Stenberg

2003

Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

The assembly of species into communities and the importance of environmental filtering, species interactions and stochastic factors in that process is one of the central topics in plant community ecology. Several studies indicate that assembly of plant communities is mostly deterministic, governed by environmental filtering and species interactions (Weiher & Keddy 1999; Adler 2004; Turnbull et al. 2005). Other studies suggest that neutral processes act on assembly (Hubbell 2001; Watkins & Wilson 2003) and that species occurrence is mainly constrained by the ability of a species to disperse to and establish at a site (i.e. dispersal limitation; e.g. Zobel et al. 2000; Freestone & Inouye 2006; Myers & Harms 2009).

A common approach to assembly studies is to use plant functional traits (e.g. Weiher et al. 1999; McGill et al. 2006; Garnier et al. 2007; Cornwell & Ackerly 2009; Götzenberger et al. 2012). Functional traits are thought to reflect general adaptations to variation in the environment and trade-offs among different functions within a plant (Lavorel et al. 2007). Species with similar traits are assumed to occupy similar niches, have similar functional roles and respond similarly to the environment (Lavorel et al. 2007; Violle & Jiang 2009). It has also been suggested that differences in functional traits determine the outcome of competitive interactions and species sorting into communities (e.g. Keddy 1992; Díaz et al. 1998; Lavorel et al. 2007; Violle et al. 2009).

Semi-natural grasslands are among the most species-rich habitats in Northern Europe (e.g. Kull & Zobel 1991; Cousins & Eriksson 2002). These grasslands usually have a long history of grazing and mowing (Eriksson et al. 2002), and have traditionally not been ploughed or fertilized. The high species richness is thought to be an effect of large historical areas, management continuity, constant disturbance by grazing animals, low soil fertility (especially phosphorus) and propagule pressure from the region (Janssens et al. 1998; Cousins & Eriksson 2002; Eriksson et al. 2002, 2006). Because of land-use changes during the past century, species-rich grasslands have declined drastically in area and are today few and often small (Eriksson et al. 2002). Grasslands that have developed on grazed ex-arable fields often occur in the same agricultural landscape as semi-natural grasslands but are of a much younger age, at an earlier succession stage and are also comparatively species-poor (Cousins & Eriksson 2002). In contrast to semi-natural grasslands, the species composition of ex-arable fields is not linked to previous management in the area. Thus semi-natural grasslands and ex-arable fields are ideal habitats for studying community assembly, where the two grassland types represent similar ecosystems in the same landscape, but differing in historical legacy, successional stage and species composition and richness. Several studies have found that natural colonization on ex-arable fields takes a long time and may be contingent on soil fertility and distance to appropriate seed sources (e.g. Pywell et al. 2002; Cousins & Aggemyr 2008; Cousins & Lindborg 2008). However, few studies have investigated trait assembly in those communities (e.g. Öster et al. 2009a).

The overall objective of this study was to investigate how plant functional trait diversity and environmental factors influenced community assembly in species-poor ex-arable fields and species-rich semi-natural grasslands. We also explored if the difference in species richness between the two grassland types was connected to variations in those factors. We incorporated traits related to both the established phase of the plant's life cycle (specific leaf area, SLA, and leaf dry matter content, LDMC) and the establishment phase (seed mass). Species traits were also measured separately for each site, instead of assuming that species traits were constant over the environment.

For our analysis, we used among other a trait gradient analysis that decomposes species trait values into alpha (within-community) and beta (among-community) components (Ackerly & Cornwell 2007). Dispersion of traits within communities may show random distribution (e.g. Watkins & Wilson 2003), divergence or convergence (Grime 2006). If the measured traits have no influence on the probabilities of plants to disperse, survive and reproduce in a certain environment, then the trait assembly at a site should be a random sample from the regional species pool (Shipley 2009). If resource competition, and thus limiting similarity, is influencing the community assembly, we would expect to see evidence of trait divergence (e.g. Weiher & Keddy 1995; Weiher et al. 1998), as competition is stronger between species that are more similar in their resource use and functional traits (MacArthur & Levins 1967; Pacala & Tilman 1994; De Bello et al. 2009; Mayfield & Levine 2010). Therefore, only species with divergent functional traits are able to co-exist. However, some authors have suggested that if species are similar enough they can escape this rule of limiting similarity and co-exist (Scheffer & van Nes 2006; Yan et al. 2012). Trait convergence (under-dispersion) often indicates environmental filtering, as only species from the species pool with certain traits can tolerate the site conditions and are able to establish and grow there (e.g. Keddy 1992; Weiher & Keddy 1995). A complicating factor is that competition can also cause trait convergence, for example when only tall species may survive under conditions of intense competition for light (Mayfield & Levine 2010). In addition, if trait convergence caused by environmental filtering, and trait divergence caused by species interactions, are working simultaneously they could also cancel each other out, giving the impression of random assembly. However, by using an approach that decomposes species trait values into alpha (within-community) and beta (among-community) components will allow us to separate the influence of environmental filtering from species interactions. Thus, trait convergence, causing higher or lower species beta values than expected at random, would imply environmental filtering.

We also applied a fourth-corner analysis (Legendre et al. 1997; Dray & Legendre 2008) to the data, which tests the null hypothesis that species and their associated traits are randomly allocated into sites with respect to measured environmental conditions, and thus whether environmental filtering is influencing the species trait assembly. While the trait gradient analysis explores variation in trait values within and among communities to determine if environmental filtering is influencing the species assembly, the fourth-corner analysis directly links species trait assembly to measured environmental factors. Furthermore, if we see some relationship between measured environmental factors and species traits/richness/abundance, it indicates some kind of environmental filtering influencing the species assembly. If species are sorted into the two grassland types according to species traits or differences in environmental factors, we would expect to see differences in average values between the two grassland types.

Previous studies have shown that species presence and species abundance are not always influenced by the same factors, therefore the data were analysed for both weighted averages (considering species abundance) and species occurrence (considering only species presence; Cingolani et al. 2007).

The three functional traits chosen are only a fraction of all possible measurable traits, and studies have shown that different traits show different assembly patterns (e.g. Watkins & Wilson 2003; Valladares et al. 2008). Therefore, the ability of this study to demonstrate the effect of species functional trait diversity on community assembly is limited, given the small number of traits. However, the results were expected to provide an insight into the relationship between the measured functional traits and co-existence, as well as the mechanism that controls functional diversity and assembly within and between the target communities.

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

Study site

The study took place in and near Nynäs nature reserve, southeast Sweden (58°49′ N, 17°24′ W; Cousins & Eriksson 2002). The study sites were species-rich semi-natural grasslands (eight sites) and species-poor ex-arable fields (six sites), located within a 3 km × 4 km large rural area. The area is a cultivated landscape with a mosaic of arable fields, ex-arable fields, semi-natural grasslands and forests. All the sites were grazed with livestock (sheep or cattle). Based on aerial photographs from 1945 and 1981 and a field survey in 1998 (S.A.O. Cousins, unpubl. data), the age of the ex-arable fields was estimated at between 20 and 40 yr.

Abundance measurements

At each of the 14 sites, five plots (2 m × 1 m) were established in a homogeneous, flat and stone-free area of ca. 10 m × 10 m, at least 5 m from the field edge. Within these plots, eight 0.3 m × 0.3 m quadrats were laid out. In these quadrats, cover of each vascular plant species was estimated from 0% to 100%. This cover measurement was then used as an estimate of species abundance. Species plot abundance was then estimated as the average species abundance from the eight 0.3 m × 0.3 m quadrats within a plot, and plot species richness as the number of species found in at least one quadrat within a plot (Fig. 1). Abundance measurements were conducted during the summers 2007–2008.

image

Figure 1. At each of the 14 grassland study sites, five plots (2 m × 1 m) were established in a homogeneous, flat and stone-free area of ca. 10 m × 10 m. Within these plots, eight 0.3 m × 0.3 m quadrats were laid out; in these quadrats, cover of each vascular plant species was estimated from 0% to 100%.

Download figure to PowerPoint

Measurements of functional traits

We measured three plant functional traits: specific leaf area (SLA; mm2·mg−1), leaf dry matter content (LDMC; mg·g−1) and seed mass (mg). The first two traits are related to the established phase of plant life cycle, while seed mass is related to the establishment phase. SLA is the area of a fresh leaf (mm2) divided by its oven-dry mass (mg). High SLA implies thinner and less dense leaf tissue, which is often associated with higher metabolic rates per mass, shorter leaf life span and higher relative growth rate, while low SLA implies denser leaf tissue and slower metabolic rates per mass (e.g. Weiher et al. 1999; Westoby et al. 2002; Cornelissen et al. 2003). Higher SLA values are often found for species in resource-rich environments (Cornelissen et al. 2003). LDMC is the oven-dry mass (mg) of a leaf divided by its fresh mass (g). It reflects the average density of the leaf tissue. High LDMC indicates tougher leaves that are more resistant to physical hazards than leaves with a low LDMC (Cornelissen et al. 2003). Studies have shown that SLA and LDMC are inversely correlated (e.g. Weiher et al. 1999) and this was also true for the present study (Pearson's product moment correlation, r = −0.55, < 0.001). From each site where species occurred, one leaf from six individuals per species was collected for measurements of SLA and LDMC. In total, leaves from 67 species were sampled, which accounted for 92–100% (mean 97%) of the total site species abundance.

Seed mass is the dry mass (mg) of an average seed of a species. Small seeds are often produced in greater numbers than large seeds (e.g. Turnbull et al. 1999; Leishman 2001), but large seeds usually contain more resources (e.g. Eriksson 1997; Dalling & Hubbell 2002; Moles & Westoby 2004). Thus small-seeded species are expected to be better dispersers, while large-seeded species are expected to have higher establishment success. In addition, small-seeded species are thought to have a more persistent seed bank (Thompson et al. 1998), a larger seed bank (Dalling & Hubbell 2002; Jakobsson et al. 2006), faster life cycles i.e. reach maturity earlier (Moles & Westoby 2006) and faster initial growth (Turnbull et al. 2008). Seeds were collected from species bearing seed during the study period (mid-Jul to mid-Sep 2008). From each species, a total of 50–100 seeds from four to ten individuals were collected at each site having seed bearing plants of that species. In total, seeds of 44 species were collected, which accounted for 66–98% (mean 80%) of the total site species abundance.

SLA, LDMC and seed mass were measured in the lab using the guidelines in Cornelissen et al. (2003).

Soil measurements

To examine if differences among communities could be attributed to differences in environmental factors, soil samples were collected from each site in early autumn 2008. The samples were taken from the top soil below litter and vegetation. Each site sample consisted of eight smaller soil cores (2.5 cm × 10 cm) that were mixed to a gross sample and then sent to Eurofins Food & Agro (http://www.eurofins.se), an accredited laboratory, for standard soil analyses of pH, P, inline image, K, Ca, Mg and soil moisture.

Data analysis

Within each site, species site abundance, species mean abundance, mean plot species richness and site species richness were calculated. Species site abundance was the average species plot abundance from the four plots within a site. Species mean abundance was the average species site abundance at sites at which the species occurred. Mean plot species richness was the average species richness from all plots within a site. Site species richness was the total number of species found in at least one quadrat within a site.

A Wilcoxon rank sum test was used to analyse differences between semi-natural grasslands and ex-arable fields in site species richness and mean plot species richness. A multiple response permutation procedure (MRPP) with Bray–Curtis distance metrics was used to test if there was a significant difference in species composition between semi-natural grasslands and ex-arable fields. A CCA analysis was conducted to explore how species were arranged into sites, unrelated to grassland type, and how and if measured environmental variables influenced the species distribution among the 14 study sites. Soil phosphorus, ammonium, moisture and calcium were the environmental variables used. Magnesium, potassium and pH were omitted from the analysis as they were highly positively correlated with calcium. A forward selection was used to decide which environmental variables significantly affected species distribution and should be included in the final ordination, variables having > 0.05 were omitted from the final CCA.

Trait gradient analysis (Ackerly & Cornwell 2007) was applied to the abundance-weighted species trait data (SLA, LDMC and seed mass). Seed mass data were log transformed before analysis to obtain normal distribution. For this analyses, abundance-weighted species mean trait value (inline image, hereafter called species trait value) was calculated for each species, and abundance-weighted site mean trait value (inline image hereafter called site trait value) for each site;

  • display math(1)
  • display math(2)

where tij is the trait value of species i at site j, aij the abundance of species i at site j, P is the total number of sites and S is the total number of species in the study. If a site trait value was missing for a species within a site, species trait values were used in the analysis.

The analysis partitions species trait values into alpha (within-community) and beta (among-communities) components. The beta component represents the species mean location along a trait gradient:

  • display math(3)

The alpha component represents how a species trait value is related to the trait value of co-occurring species:

  • display math(4)

A null model of community assembly, developed by Ackerly & Cornwell (2007), in which species and their associated abundance and trait values were assigned to sites at random, was then used to assess whether species were aggregated into communities with other species with more similar trait values than expected at random. This model tests for the role of environmental/habitat filtering in community assembly without using any measured environmental variables. The null model maintained the number of occurrences per species, the distribution of diversity and the intra-specific distribution of both trait values and abundance within species. After species were randomly assembled into sites, the trait gradient analysis was repeated. This was done 1000 times and the results were used to calculate the 95% confidence intervals for species alpha and beta values. A species differs significantly from this random distribution if it's true alpha and beta values fall outside the 95% confidence interval. As alpha and beta values are linear decomposition of the trait value, either they can both differ significantly from the null model, or neither differs, for each species. For more information about this method, see Ackerly & Cornwell (2007).

A Pearson's product moment correlation was used to estimate the relationship between log species mean abundance and species trait values. Correlations between total site species richness and site mean trait values and environmental values were estimated with Spearman rank test, as the data did not follow a normal distribution. Differences in measured abundance-weighted site mean trait values (LDMC, SLA and log seed mass) between semi-natural grassland and ex-arable fields were explored with a Wilcoxon rank sum test.

A Wilcoxon rank sum test was also used to analyse differences between semi-natural grasslands and ex-arable fields in environmental factors. A Spearman rank test was used to examine if site trait values showed a relationship with differences in environmental values. In addition, the correlation between functional traits and environmental variables was explored using the newer version of fourth-corner analysis (Legendre et al. 1997; Dray & Legendre 2008). Fourth-corner analysis directly measures the link between an R matrix of environmental variables to an Q matrix of species traits through an L matrix of species abundance/occurrence at the sites (Dray & Legendre 2008). As all our variables are qualitative, a Pearson χ2 is computed for each pair of species traits and environmental variables. The significance of the Pearson χ2 was tested using a permutation model that permutated the cell values within the columns of L (model 1; Dray & Legendre 2008). This model tests the null hypothesis that species and their associated traits are randomly allocated into sites with respect to environmental conditions (Aubin et al. 2009). Thus, similarly to the trait gradient analysis, it tests the effect of environmental filtering on community assembly, but instead of only relying on site trait values as an indicator of sites environmental conditions, it uses measured environmental variables. The fourth-corner analysis was performed using the 44 species for which we had seed mass data and mean trait values for each species.

All of the above analyses were conducted on both abundance-weighted data and species occurrence data. When the analyses were done on species occurrence, species weights were set to 1 if the species was found at a site, but 0 if a species was missing from a site. Statistical analyses were conducted using R 2.8.1 for Windows (R Foundation for Statistical Computing, Vienna, AT; available from http://cran.r-project.org) with the additional packages Vegan, ade4 and MASS.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

Species richness and abundance

In total, 107 species were found at the 14 study sites. Of these species, nine were only found in ex-arable fields and 60 only in semi-natural grasslands (Table S1). Species richness was much higher in the semi-natural grasslands compared to the ex-arable fields (site species richness: t6.9,14 = −8.94, < 0.001; mean plot species richness: t9.6,14 = −10.33, < 0.001; Table 1, Table S2). Species composition differed significantly between the two grassland types (A = 0.22, P = 0.001), and CCA analysis (Fig. 2) indicated that ex-arable fields were more variable in species composition than semi-natural grasslands.

Table 1. Mean (±SE) site species richness, plot species richness, soil moisture (Moisture, %), soil pH (pH), ammonium (NH4+), phosphorus (P), potassium (K), magnesium (Mg) and calcium (Ca) in the soil in eight semi-natural grasslands and six ex-arable fields in Nynäs nature reserve
  Ex-arable fieldsSemi-natural grasslands
Site species richness21.5 ± 3.7858.4 ± 1.65
Mean plot species richness13.6 ± 2.1941.3 ± 1.55
Moisture (%)35.2 ± 3.5130.3 ± 1.11
pH5.8 ± 0.145.8 ± 0.05
NH4+ (mg·100 g−1)1.4 ± 0.111.2 ± 0.19
P (mg·100 g−1)2.8 ± 0.621.2 ± 0.1
K (mg·100 g−1)17.5 ± 0.8119 ± 2.28
Mg (mg·100 g−1)34.2 ± 6.7629 ± 4.25
Ca (mg·100 g−1)145.7 ± 24.11108.1 ± 11.96
image

Figure 2. Distribution of species in 14 grassland sites in Nynäs nature reserve shown in a CANACO (CCA) diagram. The six ex-arable sites are marked with triangles, the eight semi-natural grassland sites with dots, and species with crosses. Only significant environmental variables are shown: soil moisture and soil phosphorus.

Download figure to PowerPoint

Functional traits

For the weighted SLA and LDMC, average site values did not differ between the two grassland types, but ex-arable fields had significantly lower average site seed mass than species in semi-natural grasslands (W = 4.00, P = 0.008). For the occurrence data, species in ex-arable fields also had significantly lower average seed mass than species in semi-natural grasslands (W = 6.00, P = 0.02), but there was also a higher average SLA in ex-arable fields (W = 44, P = 0.008).

Species aggregation (both for abundance-weighted and occurrence data) into sites was random in relation to seed mass, SLA and LDMC, as none of the species with measured SLA and LDMC (67 species) and seed mass (44 species) had beta or alpha values significantly different from the null model. Species varied more in their trait values relative to co-occurring species (alpha values) than they did in the mean trait values of sites in which they occurred (beta values; Table 2).

Table 2. Range of trait values, beta-trait values and alpha-trait values for species specific leaf area (SLA; mm2·mg−1), leaf dry matter content (LDMC; mg·g−1) and log seed mass (mg) measured from six ex-arable fields and eight semi-natural grasslands in Nynäs nature reserve, Södermanland, Sweden
  N Trait rangeBeta rangeAlpha range
AbPrAbPrAbPr
  1. N – number of species that the range is based on, Ab – values based on abundance-weighted trait values. Pr – values based on species occurrence.

SLA6726.627.95.84.623.926.1
LDMC67318.6317.649.634.0323.8317.3
Seed Mass441.92.10.50.22.02.1

Correlation between traits and species abundance and richness

There was a positive relationship between species mean abundance and species LDMC (r = 0.29, P = 0.02), and a negative relationship between species mean abundance and species SLA (r = −0.29, P = 0.02). There was no relationship between species mean abundance and species seed mass. There was a positive relationship between species richness (both site and mean plot richness) and site seed mass. There was a negative relationship between unweighted site SLA and both mean plot and site species richness and between site species richness and abundance-weighted site LDMC (Table 3).

Table 3. Results from correlation analyses on the relationship between plant functional traits (SLA, specific leaf area; LDMC, leaf dry matter content and seed mass) and species richness (both site mean and site total species richness) from six ex-arable fields and eight semi-natural grasslands in Nynäs nature reserve, Södermanland, Sweden
 Mean plot species richnessSite species richness
AbPrAbPr
  1. N – number of species that the range is based on, Ab – values based on abundance-weighted trait values, Pr – values based on species occurrence. Values are correlation values; − or + in front of a value indicate if the relationship is positive or negative, significant results are marked with asterisks: ***<0.001, **0.001–0.01 and *0.01–0.05.

SLA+0.046−0.656*+0.159−0.559*
Seed Mass+0.715**+0.632***+0.630*+0.825***
LDMC−0.53−0.100−0.555*−0.088

Soil factors

No significant differences were found in soil factors between semi-natural grasslands and ex-arable fields, but there was a tendency towards higher phosphorus levels in ex-arable fields (W = 39, P = 0.058). According to the CCA analysis, both soil phosphorus (χ2 = 0.32, P = 0.02) and soil moisture (χ2 = 0.39, P = 0.01) influenced abundance-weighted species distribution and explained 31% of the total inertia. Only soil phosphorus influenced the occurrence of species at a site (χ2 = 0.33, P = 0.01) and explained 23% of the total inertia. The only relationship between species richness and environmental variables was a negative relationship between species richness and site phosphorus levels (site species richness: r = −0.61, P = 0.02; mean plot species richness: r = −0.64, P = 0.01).

The direct relationships found between functional traits and environmental factors were a negative relationship between abundance-weighted site seed mass and both soil moisture and soil phosphorus levels, and a positive relationship between unweighted site SLA and soil phosphorus levels (Table 4). According to the fourth-corner analysis, abundance-weighted data variation in seed mass and abundance-weighted and -unweighted variation in LDMC could be linked to both variation in soil moisture and phosphorus. Variation in abundance-weighted SLA could be linked to variation in soil moisture and variation in unweighted SLA to soil phosphorus (Table 4).

Table 4. Results from fourth-corner and correlation analyses on the relationship between plant functional traits (SLA, specific leaf area; LDMC, leaf dry matter content and seed mass) and soil phosphorus and soil moisture levels from six ex-arable fields and eight semi-natural grasslands in Nynäs nature reserve, Södermanland, Sweden
 Soil FactorFourth-cornercorrelation
AbPrAbPr
  1. N – number of species that the range is based on, Ab – values based on abundance-weighted trait values, Pr – values based on species occurrence. Values are correlation values; − or + in front of a value indicate if the relationship is positive or negative; significant results are marked with asterisks; ***<0.001, **0.001–0.01 and *0.01–0.05.

SLAPhosphorus−0.123+0.101**+0.080+0.633**
Moisture−0.177*+0.041−0.257+0.278
Seed massPhosphorus−0.138**+0.005−0.628*−0.421
Moisture−0.205***−0.019−0.569*−0.169
LDMCPhosphorus+0.157*+0.072*+0.516+0.120
Moisture+0.228***+0.076*+0.120+0.054

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

We found no evidence of environmental filtering affecting the trait-based community assembly in the study systems, using the trait gradient analysis. For all traits, the range of alpha values exceeded the range of beta values, demonstrating that species varied more in trait values relative to co-occurring species than they did in the mean trait values of sites in which they occur. This indicates that most species occupied the same range of environments and therefore the underlying differences in the traits appeared as alpha trait variation. Ackerly & Cornwell (2007) suggested that if the high alpha trait differences are not associated with any mechanisms that promote co-existence, then the species are truly identical in performance (sensu Hubbell 2001). Further analyses of the species trait data and of soil factors, however, revealed that species occurrence and abundance was indeed related to both environmental factors and species traits, as well as grassland type.

Several studies focusing on many functional traits have found assembly rules only for some traits (Watkins & Wilson 2003; Valladares et al. 2008). Thus, environmental filtering might be working on other traits not measured in this study. It might also be that the traits measured are not linked to the environmental gradient among the sites. This, however, is unlikely as the fourth-corner analysis and simple correlations using the same traits revealed a relationship between environmental factors and trait assembly.

The grasslands studied cover a narrow environmental gradient, probably resulting in relatively small variation in functional traits among sites compared to other trait gradient studies (e.g. Ackerly & Cornwell 2007). Under these conditions, the trait gradient analysis might possibly have too little power to detect the influence of environmental filtering on species assembly. Fourth-corner analysis does however directly link the influence of measured environmental factors to species trait assembly and should therefore be more powerful in detecting those small differences. Similarly, comparing mean trait values between the two grassland types is probably a more successful method for revealing these differences than a method that studies community assembly among all the sites.

Other studies have also found a lack of evidence for trait-based organization of communities, e.g. for various leaf traits (Watkins & Wilson 2003), for plant height and potential above-ground biomass, and for seed mass (Schamp et al. 2008), and in a meta-analysis of 21 papers reporting results from a trait-based approach only 18% of cases reported significant deviation from the null model (Götzenberger et al. 2012). So the majority of similar studies based on traits and null models are coherent with our findings. It has, however, been implied that the failure of these studies to detect non-random assembly might simply be caused by the model design (Götzenberger et al. 2012), and this is supported by our study.

Results for other statistical methods than the trait gradient analysis revealed that species-poor sites had proportionally more species with a short generation time that produce many seeds (lower seed mass; Moles & Westoby 2004) and had higher growth rates (higher SLA; Cornelissen et al. 2003) compared to more species-rich sites. In addition, species with larger seedlings (higher seed mass; Moles & Westoby 2004) and thinner leaves (lower LDMC; Cornelissen et al. 2003) were more abundant on species-rich sites. Furthermore, both soil moisture and soil phosphorus also affected species assembly at a site. Species richness increased with decreased phosphorus levels, and fourth-corner analysis revealed that trait assembly into the grasslands was non-random in relation to both site phosphorus level and soil moisture. This begs the question if the link between species richness and measured traits is a direct link related to species seed production and/or competitive abilities, or a link mediated by the relationship between species traits and environmental factors. Other studies have shown that, on average, perennial species in nutrient-poor and/or dry habitats have thicker or tougher leaves than those occurring in more resource-rich habitats (Fonseca et al. 2000; Niinemets 2001; Wright et al. 2002; Cornwell & Ackerly 2009), and higher SLA values (thinner leaves) are often found for species in resource-rich environments (Cornelissen et al. 2003). This is only partly consistent with our results, which found that species with lower SLA (thus thicker leaves) were more common on sites with low phosphorus levels and more abundant on wetter sites. Species with higher LDMC (tougher leaves) were associated with wetter site and sites with higher phosphorus levels. These contradictory results indicate that perhaps the variation in SLA and LDMC is not directly linked to variation in environmental factors but rather mediated by some other factors. In addition, at resource-rich sites (sites with high soil phosphorus and moisture), species that mature quickly and produce many seeds were more abundant, while in resource-poor environments species with large seeds and thus higher establishment rates (Leishman & Westoby 1994) were more abundant.

To summarize, in the studied grasslands, as species richness increases, species dominance shifts from species that can quickly establish at a new site to species that produce stronger seedlings. Whether this shift is mostly related to increased competition, dispersal limitation or to changes in environmental factors, is hard to determine.

Differences between the two grassland types

There was a significant difference in species composition and species richness between the two grassland types. Semi-natural grasslands had much higher species richness, and the species composition was more similar among different semi-natural grassland sites than among ex-arable fields. Semi-natural grasslands had higher average seed mass, both for occurrence and abundance-weighted data, implying that large-seeded species were lacking in ex-arable fields. We postulate that the lack of large-seeded species in ex-arable fields is related to the younger age and earlier successional stage of these ex-arable fields. This is strengthened by the lack of many late-successional species in ex-arable fields, such as Anemone nemorosa, Betula pendula, Calluna vulgaris, Hieracium umbellatum, Vaccinium myrtillus, Scorzonera humilis, Succia pratensis and Viola palustris. In young communities, it is likely that easily dispersed species are favoured in the colonization process (Öster et al. 2009b). Species with low seed mass may be faster to establish there as they often produce more seeds and are therefore more likely to colonize the area, even though larger seeds usually have higher recruitment rates than smaller seeds in grasslands (e.g. Jakobsson & Eriksson 2000; Ozinga et al. 2005; Kahmen & Poschlod 2008). In addition, species with lower seed mass usually have a shorter generation time (Moles & Westoby 2004) and could therefore be faster to establish and to increase in abundance at young sites. Previous studies have also shown that colonization is more dependent on seed abundance than on resource content of the seed (Duarte et al. 2007). The more abundant a species is in the seed rain, the more likely it is to disperse to a site, thus overcoming the dispersal limitation. Other studies in ex-arable fields in Sweden (Öster et al. 2009a) and in the UK (Pywell et al. 2002) have also found that dispersal limitation is an important factor in explaining the low species richness found there. Dispersal is not only affected by seed abundance, but dispersal vectors, among other factors, can also influence species dispersal potentials (Cornelissen et al. 2003; Vittoz & Engler 2007), especially in fragmented landscapes. However, in our study area all species, unrelated to dispersal vector, should be able to disperse freely between the two grassland types, as the ex-arable fields in this study were all close to semi-natural grasslands, without any physical dispersal barrier between them.

When looking at traits related to the established phase of the plant life cycle, we found no differences in LDMC trait values between the two grassland types, but there was a higher occurrence of species with higher SLA in ex-arable fields, indicating that species with lower SLA were missing from ex-arable fields. Higher SLA has been linked to a higher growth rate (Cornelissen et al. 2003), which could imply that plants ‘mature’ more quickly, which is consistent with the conclusion that lower seed mass in ex-arable fields can be explained by shorter generation time of small-seeded species. Even though SLA seems to influence which species colonize ex-arable fields, it does not affect the abundance of species. This is coherent with findings from Cingolani et al. (2007), who concluded that the traits which determine the probability of species presence at a site are not necessarily the same as those which determine site abundance.

Species with higher SLA and heavier seeds often have better competitive abilities and higher establishment success (Moles & Westoby 2004), indicating that the differences observed between the two grassland types could, in addition to dispersal limitations, be caused by limitations on species establishment in ex-arable fields. If this is the main cause of the differences in species assembly between the two grassland types, it would imply that species with better competitive abilities and higher establishment success had lower establishment rates in species-poor ex-arable fields than other species. We found no differences in measured environmental factors between the two types of grassland that could explain these differences, which is consistent with other studies in the region (Öster et al. 2009a). There was, however, a tendency towards ex-arable fields having higher soil phosphorus levels than semi-natural grasslands, but the phosphorus levels were usually below 5 mg·100 g−1 soil, which has been suggested as a level at which phosphorus starts to severely reduce recruitment in grassland (Janssens et al. 1998). Phosphorus levels were negatively correlated with average seed mass. High soil phosphorus levels in ex-arable fields might thus influence which species can establish there. Furthermore, additional soil factors, not measured in this study, may differ between the two grassland types, thus influencing species establishment. Other studies have however concluded that seed limitation, rather than high soil fertility, is the main constraint on species colonization in ex-arable fields (Pywell et al. 2002).

Thus, even though we imply that differences between the two grassland types is mainly caused by dispersal limitations, we cannot exclude that the differences are also related to establishment limitations, caused by e.g. differences in environmental factors or species competition not captured in this study.

Conclusions

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

The assembly of plants species is a complex process. While the trait gradient analysis strengthens the view that assembly of communities from a pool of suitable colonizers is highly influenced by random factors (e.g. Gleason 1926; Foster et al. 2004; Vandvik & Goldberg 2006), other results indicate that assembly is not random but highly influenced both by soil factors and species traits. The limitations of using models to determine species assembly have been discussed previously (Van der Maarel et al. 1995; Gotelli 2000; Götzenberger et al. 2012), and these limitations are obvious in the current study. In addition, this study underlines the importance of age of the communities in community assembly, showing that small-seeded species are quicker to establish at new sites than species with larger seeds, while at older sites species with both large and small seeds have become established.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

We are grateful to H. Rydin, M. Öster, K. Hylander and three anonymous reviewers for valuable comments on earlier drafts of this manuscript. We also thank M. Fjäder, C. Niklasson and M. Enskog for help with field studies and in the laboratory. This study was financially supported by the Swedish Research Council.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information
FilenameFormatSizeDescription
jvs12058-sup-0001-TableS1.pdfapplication/PDF12KTable S1. Species list of all species found at the 14 study sites.
jvs12058-sup-0003-TableS2.pdfapplication/PDF29KTable S2. Species richness and measured soil factors at the 14 study sites.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.