Interspecific differences in determinants of plant species distribution and the relationships with functional traits

Authors

  • Masahiro Aiba,

    Corresponding author
      Correspondence author. E-mail: mshiro5@gmail.com
    Search for more papers by this author
    • Present address: Graduate School of Life Sciences, Tohoku University, Aoba 6-3, Aramaki, Aoba-ku, Sendai 980-8578, Japan.

  • Hino Takafumi,

    1. Tomakomai Research Station, Field Science Center for Northern Biosphere, Hokkaido University, Takaoka, Tomakomai 053-0035, Japan
    Search for more papers by this author
  • Tsutom Hiura

    1. Tomakomai Research Station, Field Science Center for Northern Biosphere, Hokkaido University, Takaoka, Tomakomai 053-0035, Japan
    Search for more papers by this author

Correspondence author. E-mail: mshiro5@gmail.com

Summary

1. Environmental control and dispersal limitation are both essential processes in plant community assembly and species distribution. Although numerous studies in the past decade have examined their importance as determinants of community composition, remarkably little is known about interspecific differences in the importance of these two processes.

2. To quantify these interspecific differences, we compared the importance of environmental factors and space as correlates of species distribution among 24 understorey plant species in a Japanese cool–temperate forest by performing variation partitioning at the species level. Specifically, we hypothesized that the importance of environment and space differs among species, and these differences can be partly predicted from the functional traits and/or phylogenetic identity of each species.

3. The unique contributions of both environment and space were significant in the community-level analysis. However, at the species level, the relative and absolute sizes of the unique contributions of environment and space differed considerably among the 24 species. Environment and space were not necessarily significant variables explaining the distribution of many species.

4. No significant relationships were found between the unique contribution of environment and the four functional traits tested, that is, dispersal mode, seed mass, plant height and specific leaf area among the 24 species. In contrast, the unique contribution of space was significantly larger in species with no dispersal mechanisms than in animal-dispersed species. No significant phylogenetic signal was detected for the unique contribution of environment or space, suggesting that importance of environmental control and dispersal limitation as determinants of species distribution is evolutionarily labile.

5. Synthesis. Our results suggest that the relative and absolute importance of different processes of community assembly (i.e. environmental control and dispersal limitation) differs remarkably among species even within a single community. These interspecific differences may be explained in part by interspecific differences in dispersal mode.

Introduction

Understanding the assembly processes of a community is one of the central themes of ecology. Recent studies have demonstrated that both environmental control and dispersal limitation play essential roles in assembly processes (Cottenie 2005; Myers & Harms 2009). Biotic and abiotic environmental factors control the abundance of species via their effects on demographic traits, that is, survival, growth and reproduction (Hutchinson 1957; Grubb 1977; Nathan & Muller-Landau 2000; Pulliam 2000). If environmental control dominates assembly processes, community composition and species distribution are expected to be rather deterministically predictable by understanding species’ environmental preferences and competitive abilities, as well as the environmental heterogeneity of the site. On the other hand, the spatio-temporal shortage of dispersal units excludes species from their potential habitat, regardless of the environment (Nee & May 1992; Tilman 1994; Pulliam 2000; Hubbell 2001; Calcagno et al. 2006). As a result, if dispersal limitation is a dominant assembly process, community composition and species distribution would be spatially structured independent of environment. The relative importance of these two processes (i.e. environmental control and dispersal limitation) in plant community assembly has been examined in numerous studies in the last decade (e.g. Gilbert & Lechowicz 2004; Svenning et al. 2004; Cottenie 2005; Clark et al. 2007; Myers & Harms 2009).

The statistical procedure of variation partitioning has been used as an effective tool in these studies (Borcard, Legendre & Drapeau 1992). Variation partitioning divides the total variance of a response variable (here, community data) into two or more subsets, which are respectively explained by suites of explanatory variables (here, environmental and spatial variables). Ordination techniques such as redundancy analysis (RDA) or canonical correspondence analysis (CCA) have been used to relate the variance of a response variable with explanatory variables. For plant communities, studies using variation partitioning framework have demonstrated that both environment and space are moderately important as determinants of the spatial pattern, from highly diverse tropical regions to less species-rich areas at higher latitudes (e.g. Gilbert & Lechowicz 2004; Svenning et al. 2004; Karst, Gilbert & Lechowicz 2005; Jones et al. 2008; Legendre et al. 2009; Flinn et al. 2010). For example, in the case of a Canadian wetland herb community, the unique contributions of environment, space and the spatially structured environment (i.e. variance shared by environment and space) made up 9.7%, 9.1% and 13.4% of the total variance, respectively (Flinn et al. 2010). Similarly, for a Costa Rican pteridophyte community, the three components were 17%, 6% and 9%, respectively (Jones et al. 2008). However, our understanding of interspecific differences in the importance of environment and space as determinants of spatial distribution is still remarkably limited, as all analyses have been performed at a community level. This is surprising given that variation partitioning at the species level is not technically difficult, as the explained variance at the community level offered by RDA is a weighted mean of the R2 of multiple regressions for each of the constituent species (Peres-Neto et al. 2006).

An interspecific comparison of the importance of environment and space as determinants of spatial distribution is essential for answering several ecologically important questions. First, two major theories based on dispersal limitation, the neutral model and the competition–colonization trade-off hypothesis, operate under contrasting assumptions for interspecific differences in the extent of dispersal limitation. Basic neutral models assume that both dispersal ability and the probability of local extinction are uniform among species (Bell 2000; Hubbell 2001). On the other hand, interspecific differences in dispersal ability are the essence of species coexistence in the competition–colonization trade-off hypothesis (Nee & May 1992; Tilman 1994; Calcagno et al. 2006). Second, despite accumulating evidence of interspecific differences in plant dispersal ability (see a review by Vittoz & Engler 2007), whether such differences consequently lead to interspecific differences in the extent of dispersal limitation remains ambiguous (Flinn et al. 2010).

Once interspecific differences in the contribution of environment and space to species distribution are detected, seeking links between these contributions and functional traits of a species would be an interesting next step. Recently, Flinn et al. (2010) performed variation partitioning for a subset of a wetland herb community, which was grouped by dispersal mode, to show that spatial variables are more important for a group of species with limited dispersal ability than for a group of species with higher dispersal ability. Several studies on aquatic organisms have reported similar results (Beisner et al. 2006; Van De Meutter, De Meester & Stoks 2007; Vanschoenwinkel et al. 2007). However, the methodology of these studies is not appropriate for relating the results of variation partitioning with non-categorical traits. Many essential functional traits of plants, for example seed mass, plant height and specific leaf area (SLA) (e.g. Westoby 1998), are continuous variables. In addition to seed traits, plant height is an important determinant of dispersal distance (Soons et al. 2004; Thomson et al. 2011). Vegetative traits such as SLA may also relate to the strength of dispersal limitation by changing the probability of establishment after dispersal (Tremlova & Munzbergova 2007). Species with certain traits may also be more severely controlled by environment. For example, Cornwell & Ackerly (2009), who analysed trait distribution patterns of trees in a California forest, found that trait ranges of wood density and tree height were positively correlated with soil moisture because the occurrence of short-shrub species with dense wood was confined to wetter sites. In this case, the strength of environmental control may be positively correlated with wood density and negatively correlated with plant height within the community. Furthermore, analysis at the species level provides an opportunity to test the strength of phylogenetic signals (Blomberg, Garland & Ives 2003; Losos 2008) in determinants of species distribution. Phylogenetic signals may be detected even if we fail to find correlations between functional traits and the strength of environmental control or dispersal limitation, which would provide evidence suggesting the importance of untested traits.

In this study, we comparatively analysed the importance of environment and space as correlates of species distribution of understorey plants in a Japanese cool–temperate forest. Our hypotheses were that the importance of environment and space differ among species and that interspecific differences can be predicted from the functional traits and/or phylogenetic identity of each species. In particular, we expected a higher contribution of space in species with no dispersal mechanisms (gravity-dispersed species), large seeds and/or shorter heights, due to the limited dispersal ability of these species.

Materials and methods

Study Site

The study was conducted in the 2715-ha Tomakomai Experimental Forest (TOEF), located in Hokkaido, the northernmost main island of Japan (42°41′N, 141°36′E). The mean monthly temperature is 6.7 °C, with the highest monthly mean of 19.9 °C in August and lowest of −6.1 °C in January. Mean annual precipitation is 1100 mm, and snow cover reaches a depth of 50 cm from December to March. A large part of TOEF is flat and the inclination is <5°. The forest established on 2-m deep regosols accumulated during the eruptions of a nearby volcano, Mt. Tarumae, in 1669 and 1739. Approximately 350 vascular plants have been recorded in TOEF (Kudo & Yoshimi 1916). The dominant canopy tree species in the natural stands are Quercus crispula, Acer mono, Sorbus alnifolia and Tilia japonica (Hiura 2001). The current landscape of TOEF is a mosaic of primary forest, secondary forest and plantations of various tree species; each stand type occupies one-third of TOEF (Hiura 2005).

Data Collection

We randomly located 60 square quadrats of 9 m2 in primary stands. We recorded coverage, to the nearest 10%, for each species of non-woody vascular plants, lianas and small shrubs, which were typically <1 m in height, in June and July 2010. Coverage was estimated by two independent observers, and mean values were used for analyses. Juveniles under 5 cm were excluded because identification was often difficult among related species, but adults of very small species that often reach maturity when <5 cm tall were included. Two species of Trillium were excluded from analyses as reliable identification was difficult without flowers. In total, 96 species were included in the community-level analysis. For variation partitioning at the species level, we focused on the abundance distribution of 24 relatively frequently occurring species that were present in at least 20 quadrats. The 24 species are listed in Table 1.

Table 1.   Unique contributions (percentage) of the six selected environmental variables to the spatial abundance distribution of the 24 species studied
GenusSpeciesSubspecies/variationSouth-facing slope inline image Mg (quadratic)West-facing slopeSlope angleSoil humus content
  1. Signs for the contribution of environmental variables represent direction of the effects. Values with asterisks were significant at P < 0.05 after 999 permutations.

Adoxa moschatellina  −12.3*00−7.1*1.50
Carex japonica  −0.10−2.800−1.0
Carex rugata  001.2000
Chamaele decumbens  05.4*−0.10−7.3*0
Chloranthus serratus  0000.10−0.3
Daphne pseudo-mezereum subsp. jezoensis −5.7*0−0.100−4.6*
Diarrhena japonica  00−1.7000
Dryopteris crassirhizoma  −5.1*1.2−17.6*000
Galium japonicum  00000−1.0
Galium trifloriforme  −1.300−0.2−5.4*−4.5*
Hydrangea petiolaris  −15.5*0−1.4−19.1*0−0.1
Lycopodium serratum var. serratum −3.2−5.4*001.60
Maianthemum dilatatum  2.6−14.9*−0.4000
Maianthemum japonica  −8.0*0.3006.0*0
Pachysandra terminalis  0000−3.90
Phryma leptostachya subsp. asiatica 00−6.0*0−2.60.8
Platanthera ussuriensis  00−0.60−8.5*0
Rhus ambigua  00−12.7*0−0.40
Sanicula chinensis  00.8−0.10−0.30
Schisandra chinensis  −1.8000−0.10.2
Schizophragma hydrangeoides  0−6.2*−1.10−0.10
Scutellaria indica  00.80000
Solidago virgaurea subsp. asiatica 2.2−2.20001.3
Viola selkirkii  0002.8−0.10

We used environmental variables measured in 1-m2 permanent quadrats established for a separate ongoing study adjacent to our census quadrats. The appearances of measurement points (e.g. vegetation, topography and canopy openness) were similar to those of the census quadrats in all cases. Topographic positions of each of these quadrats (hereafter in this paragraph, the term ‘quadrats’ indicates 1-m2 permanent quadrats) were classified into one of the four categories: flat, ridge, slope and valley. Aspect was defined as one of the following categories: flat (slope angle ≤5°), north, east, south and west. Topographic position and aspect were coded as dummy variables. We treated ‘flat’ as a baseline category for both topographic position and aspect. Slope angle was measured in the steepest direction across quadrats using a clinometer. Soil water content was measured at three points for each quadrat using TRIME-FM (IMKO GmbH, Ettlingen, Germany) and then averaged. The A0 layer of soil within a 0.0625-m2 frame was collected after leaf fall in autumn and oven-dried to constant weight as an index of litter accumulation. Four 5-cm deep soil samples were collected at each quadrat. These four samples were pooled and analysed for pH, inline image, P, K, Ca, Mg and humus content at JAHT Co., Ltd (Toyonaka, Japan). We took hemispherical photographs from a height of 1 m at one corner of each quadrat to measure the light environment in August 2008. Fractions of total transmitted radiation were calculated from the photographs using Gap Light Analyzer software (version 2.0; Frazer, Canham & Lertzman 1999). Current-year shoots of all vascular plants that were <1.5 m tall were harvested in a 0.25-m2 frame at a location adjacent to the quadrats in August 2009. This biomass measurement was used as an indicator of productivity. Trees larger than 5 cm in diameter at breast height (DBH) within a 5 m radius from the centre of the quadrats were measured to obtain median DBH and basal area. For continuous variables (excluding pH), we log-transformed the values assuming that unit differences of these values would be more important when their absolute sizes are small (Jones et al. 2008). We calculated variance inflation factors of the environmental variables; as a result, a dummy variable that represents ‘slope’ topography was excluded from later analyses to reduce multicollinearity. We generated quadratics for continuous variables by centring and squaring to model unimodal (or U-shaped) responses of species along environmental gradients (Flinn et al. 2010; Gilbert & Bennett 2010). Thus, our full model consisted of 37 environmental variables in total.

We generated spatial variables using the principal coordinates of neighbour matrices (PCNM) method to characterize spatial structure at multiple scales (Borcard & Legendre 2002; Dray, Legendre & Peres-Neto 2006). The PCNM variables are a suite of orthogonal variables, which is ordered based on the spatial scales they represent. To obtain the PCNM variables, we first generated a Euclidean distance matrix of the 60 quadrats. This matrix was then truncated using a threshold value (the length of the longest edge of the minimum spanning tree connecting the quadrats), above which all distances were considered very distant. (We actually replaced the real distances by four times the length of the threshold value.) Finally, we calculated principal coordinates of this matrix. We used 16 variables with positive Moran’s I values whose wavelengths ranged from about 700 to 7000 m as spatial explanatory variables (Dray, Legendre & Peres-Neto 2006).

We collected values of functional traits of the 24 species using field and literature surveys. In August 2009, plant height to the highest living part including reproductive organs of the largest individual of each species was measured (within 20 m from the centre of the 1-m2 permanent quadrats) in up to three quadrats (depending on their availability) where the abundance of the species was highest in a preliminary vegetation census performed in July 2009. Ramets were then sampled and stored in a cooler until returned to the field station. To obtain SLA, 1–20 typical-sized, sound and mature leaves, including petioles and rachises of compound leaves (analysed leaf number was depending on leaf size and availability), were digitally scanned before oven-drying (at 60 °C to constant weight). We used more than six leaves in total for most species, excluding Maianthemum dilatatum, whose ramets often consist of a single leaf. For two of the 24 species, the plant body had already senesced by the sampling period, and thus these traits were unavailable. Shoot height was also not measured for the four liana species. Height and SLA were averaged for each species for use in later analyses. Thus, these values represent average functional traits of sound individuals at relatively preferred habitats in this landscape for each species. Seed mass data were obtained mainly from Nakayama, Inokuchi & Minamitani (2000), and the Seed Information Database (Royal Botanic Gardens Kew 2008) was referenced supplementarily. Seed mass was not available in the literature or the data base for three species. Pteridophytes were excluded from the analysis of seed mass. Dispersal mode was obtained from Asano & Kuwahara (1990) and Asano (2005) and classified into one of the following categories: spore (pteridophytes), no dispersal mechanisms (gravity-dispersed species), animal-dispersed species and wind-dispersed species. When records of dispersal mode were not available in the literature, they were complemented by field observations. Functional traits of the 24 species are summarized in Table S1 in Supporting Information.

Statistical Analysis

We performed variation partitioning (Borcard, Legendre & Drapeau 1992) to quantify the contribution of the environmental and spatial variables described earlier to community structure and species distribution. While variation partitioning is usually performed at the community level using ordination methods such as RDA or CCA, in this study, we also performed variation partitioning at the species level based on multiple regression. This approach is an application of variation partitioning based on RDA, as the result of variation partitioning based on RDA is the weighted means of R2 of multiple regressions for each of the constituent species (Peres-Neto et al. 2006). We used values of adjusted R2 as explained variance because normal R2 are strongly affected by the number of samples and explanatory variables (Peres-Neto et al. 2006). First, we separately performed forward selection for environmental and spatial explanatory variables to ensure only significant variables were used in the final models. To avoid overestimation of adjusted R2, forward selection of variables was performed using adjusted R2 of the full model as a second requisite, in addition to the significance of each variable, to stop selection (Blanchet, Legendre & Borcard 2008). This step was only performed at the community level, and the same suite of variables was used for all species in later regressions at the species level to minimize the risk of selecting superfluous variables through repetition of variable selection. We then performed three multiple regressions for each species to obtain the percentage of variance explained by environmental variables only, spatial variables only and both. Finally, total variances of species abundances in our quadrats were divided into four fractions, that is, the unique contribution of environment (variance explained by environment independent of space), the unique contribution of space (variance explained by space independent of environment), the contribution of spatially structured environment (variance shared by environment and space) and residuals by sequential subtractions. These fractions can be negative, and in such cases, the fractions were bounded to zero. Additionally, we calculated the unique contributions of each environmental variable using other environmental variables and spatial variables as covariables. These unique contributions of single environmental variables occasionally exceeded the total unique contribution of the environment because some fractions were negative, as described above. The significance of the explained variances was verified using permutation tests. We also performed variation partitioning at the community level including all 96 species based on RDA. In this case, a community data matrix was Hellinger-transformed before variation partitioning (Legendre & Gallagher 2001). Later procedures were conducted in the same manner as those of the species-level analysis.

Relationships between the unique contributions of environment and space and the functional traits were tested using the Kruskal–Wallis rank-sum test (for dispersal mode) or Spearman’s rank correlation (for the other traits). For the analysis of phylogenetic signals, we constructed a phylogenetic tree of the 23 species based on the most recent phylogenetic supertree of angiosperms (R20100318, available at http://svn.phylodiversity.net/tot/megatrees/). Lycopodium serratum was excluded from phylogenetic analyses because the supertree does not include Lycopodiophyta. The branch lengths were then calculated based on known node ages using the BLADJ algorithm, which is offered in the software Phylocom (version 4.1; Webb, Ackerly & Kembel 2008). One of the indices that represents the extent of the phylogenetic signal, the K statistic (Blomberg, Garland & Ives 2003), was calculated based on the phylogeny. A K-value of 1 indicates that the traits evolved under Brownian motion, K < 1 indicates random or divergent trait evolution more than expected under the Brownian motion model and K > 1 indicates conserved patterns of trait evolution more than expected under Brownian motion. Significance was assessed by comparing the observed K statistics with the distribution of K statistics obtained by 999 permutations of trait values across tips of the tree. All statistical analyses were performed in the statistical environment r 2.12.1 (R Development Core Team 2010).

Results

In variation partitioning at the community level including all 96 species, the unique contributions of environment and space and the contribution of spatially structured environment were 7.2%, 4.0% and 1.4%, respectively. Selected environmental variables were, in order of unique contribution, south-facing slope (the unique contribution was 1.6%), soil nitrate content (1.3%), the quadratic of soil Mg content (0.87%), west-facing slope (0.82%), slope angle (0.73%) and soil humus content (0.14%). The selected spatial variables were the first, second, third, fourth, fifth and sixth of the 16 PCNM variables, which represent relatively broad-scale spatial structure.

In variation partitioning at the species level, the unique contribution of environment varied substantially among the 24 species and ranged from 0.0% to 27.7% (mean ± SD, 8.8 ± 9.1%; Fig. 1). The unique contribution of space ranged from 0.0% to 25.7% (6.3 ± 7.2%), and the contribution of spatially structured environment ranged from 0.0% to 16.5% (5.1 ± 4.8%). The relative unique contributions of environment and space against total explained variance ranged from 0.0% to 93.5% and from 0.0% to 90.1%, respectively. The unique contribution of environment and space was significant for 10 and 9 species, respectively. The abundance distributions of only 5 of the 24 species were significantly correlated with both environment and space. The contributions of the respective environmental variables were often zero after adjustment at the species level (Table 1). The effects of south-facing slope, soil nitrate content, the quadratic of soil Mg content, west-facing slope, slope angle and soil humus content were non-zero for 11, 9, 13, 5, 13 and 9 species, respectively. The directions (positive or negative) of the effects of the six selected environmental variables were rather variable among species (Table 1). The effects of south-facing slope, soil nitrate content, west-facing slope, slope angle and soil humus content were positive for 3, 5, 2, 3 and 3 species, respectively. For soil Mg content, the responses of 12 species were unimodal, whereas that of one species was U-shaped.

Figure 1.

 Variation partitioning at the species level for spatial abundance distribution of the 24 relatively frequently occurring species. Fractions that were significant at < 0.05 after 999 permutations are marked with asterisks. Note that the significance of the correlated effects of environment and space is not testable.

Correlations between the unique contribution of environment and functional traits (i.e. dispersal mode, seed mass, plant height and SLA) were generally weak and non-significant (Fig. 2). The unique contribution of spatial variables was significantly larger in species with no dispersal mechanisms than in animal-dispersed species (Fig. 3a). The sample sizes of the other two dispersal modes (i.e. spore and wind-dispersed species) were too small to evaluate the results. No significant correlations were found between the unique contribution of space and the other three functional traits (Fig. 3b–d). No significant phylogenetic signal was detected for either the unique contribution of environment (= 0.33, P = 0.37) or that of space (K = 0.18, P = 0.86) among the 23 species.

Figure 2.

 Correlations between (a) dispersal mode, (b) seed mass, (c) plant height and (d) specific leaf area (SLA) versus the unique contribution of environment. Results of Kruskal–Wallis rank-sum test (for dispersal mode) or Spearman’s rank correlation (for the other three traits) are also shown. Note that the K value in this figure is not the K statistic for phylogenetic signal but is the test statistic of the Kruskal–Wallis test. Boxes and whiskers indicate the interquartile range and maximum and minimum values within 1.5 times the interquartile range, respectively. Individual values are also plotted. Gravity, animal and wind represent no dispersal mechanisms (gravity dispersal), animal dispersal and wind dispersal, respectively. X-axes are log-scaled for continuous traits.

Figure 3.

 Correlations between (a) dispersal mode, (b) seed mass, (c) plant height and (d) specific leaf area (SLA) versus the unique contribution of space. Results of the Kruskal–Wallis rank-sum test (for dispersal mode) or Spearman’s rank correlation (for the other three traits) are also shown. Note that the K value in the figure is not the K statistic for the phylogenetic signal but is the test statistic of the Kruskal–Wallis test. Boxes and whiskers indicate the interquartile range and maximum and minimum values within 1.5 times the interquartile range, respectively. Individual values are also plotted. Gravity, animal and wind represent no dispersal mechanisms (gravity dispersal), animal dispersal and wind dispersal, respectively. X-axes are log-scaled for continuous traits.

Discussion

The relative and absolute sizes of the unique contributions of environment and space were considerably different among the 24 relatively frequently occurring plant species. For many species, both environment and space were not necessarily significant as explanatory variables of species distribution, whereas both environment and space were significant at the community level. Although these analyses were the first trial of variation partitioning of distributions at the species level, several previous studies have reported results consistent with ours. Flinn et al. (2010) reported differences in the effects of environment and space on community structure among subsets of a Canadian wetland herb community that were grouped by dispersal mode. Similarly, seed-sowing experiments quantifying the extent of dispersal limitation in mainly patchy, discrete habitats have demonstrated that the severity of dispersal limitation differs considerably among co-occurring species (e.g. Ehrlén & Eriksson 2000; Svenning & Wright 2005; Moore & Elmendorf 2006). These results suggest that interspecific differences in the importance of environment and space as determinants of distribution are widespread among plant communities.

The unique contribution of the respective environmental variables and the direction of the effects differed widely among species. These results demonstrate considerable interspecific differences in environmental preferences among the study species. Gilbert & Lechowicz (2004) showed that the relative importance of environmental variables differed among taxa or growth types in a Canadian understorey plant community. However, our species-level analysis indicated that similarity in the unique contribution of environmental variables or in the direction of those effects was relatively rare among related species, for example two Carex species, two Galium species and two Maianthemum species. Therefore, phylogenetic signals in environmental preferences appear to be weak in these species, in contrast to the results of Gilbert & Lechowicz (2004).

As we hypothesized, the unique contribution of space was significantly larger in species with no dispersal mechanisms (gravity-dispersed species) compared with animal-dispersed species. This result is not surprising, given the considerable differences in dispersal ability between species with no dispersal mechanisms and animal-dispersed species. In their review, Vittoz & Engler (2007) showed that the upper limit of the distance within which 99% of seeds will disperse is only 5 m for species with no dispersal mechanisms, whereas that for some animal-dispersed species exceeds 500 m. Due to this limited dispersal ability, species with no dispersal mechanisms would fail to reach some distant parts of their potential habitat, especially those newly created from environmental changes or local extinction. As a result, spatial distances explain a larger portion of the variance in the spatial abundance distribution of species with no dispersal mechanisms than those of animal-dispersed species.

In contrast, interspecific differences in the unique contributions of environment and space were generally independent of the other three functional traits, that is, seed mass, plant height and SLA. The lack of a linkage between the unique contribution of space and seed mass is especially interesting, as this result deviates from the assumption of the competition–colonization trade-off hypothesis of species coexistence (Nee & May 1992; Tilman 1994; Calcagno et al. 2006). According to this hypothesis, small-seeded species should be less dispersal-limited to allow escape from competition with large-seeded, often competitive species for coexistence at a metacommunity level. The absence of links between the contribution of space and seed mass is possibly due to the rather weak correlation between seed mass and dispersal distance (Thomson et al. 2011).

The results of the previous studies on the relationship between functional traits and the extent of dispersal limitation have not been consistent. In the case of Flinn et al. (2010), the contribution of space to community assembly was more important in species whose seeds are dispersed by ants, explosion, splash or gravity than in animal-dispersed or wind-dispersed species. Tremlova & Munzbergova (2007) demonstrated that patch occupancy correlates positively with wind dispersal, external animal dispersal, seed bank formation and above-ground biomass in Bohemian grasslands fragmented into agricultural fields. On the other hand, Moore & Elmendorf (2006) failed to find a correlation between the extent of seed limitation and seed mass or seed dormancy in a California grassland. In a meta-analysis of numerous seed-sowing experiments, Clark et al. (2007) showed that the extent of seed limitation is independent of dispersal mode, whereas they found a significant positive correlation between the extent of seed limitation and seed size.

In this study, interspecific differences in the unique contributions of environment and space were not explained by phylogenetic identity. In fact, the unique contributions of environment and space differed considerably even among congeneric species. For example, whereas the unique contribution of space was 10.2% for Carex japonica (P < 0.05), no unique contribution of space was detected for Carex rugata. Similarly, the unique contribution of environment was 11.7% for Maianthemum japonica (P < 0.05), but no unique contribution of environment was detected for M. dilatatum. Given that dispersal mode is often conserved among related species (Table S1), this evolutionary lability suggests that evolutionarily labile traits that were not considered in this study may play an important role in the process by which species distribution patterns are determined.

The relatively short history of the community at the study site, which began only after the explosion of a nearby volcano in 1739, may be partially responsible for the limited links between the results of variation partitioning and functional traits in this study. The study community would consist only of species whose ability to immigrate, establish and avoid local extinction has been sufficiently high to maintain a population at this site after the explosion. Such a filtering of the species pool, which may exclude strongly dispersal-limited species, would blur relationships between the unique contribution of space and functional traits. Similarly, the lack of links between the unique contribution of environment and functional traits may be attributable to the environmental homogeneity of the relatively flat landform on deep regosols. In fact, the explained variance by both environment (7.2%) and space (4.0%) at the community level at this site was relatively small compared with results of other studies using a comparable procedure (e.g. Jones et al. 2008; Flinn et al. 2010).

In conclusion, our results suggest that dispersal limitation, as well as environmental control, is an essential process in the community assembly of understorey plant species at the study site. However, at the same time, we found considerable interspecific differences in the relative and absolute importance of environmental control and dispersal limitation and of essential environmental variables that are not explicitly accounted for in basic neutral theories. These results support the importance of incorporating interspecific differences in both environmental preferences and the extent of dispersal limitation into stochastic models of community assembly, which has been attempted in several recent studies (e.g. Ruokolainen et al. 2009; Salomon, Connolly & Bode 2010). We demonstrated that interspecific differences in the importance of space as correlates of distribution were partially predictable from the dispersal mode of species. However, the results of existing reports on relationships between functional traits and the contributions of environment and space are far from consistent. A re-analysis at the species level of past studies, which were all performed at the community level, would promote our general understanding of the relationship between determinants of spatial patterns and functional traits.

Acknowledgements

We thank the staff of TOEF for assistance in the field. Angela Moles, Roberto Salguero-Gomez and two anonymous referees provided helpful comments on a previous version of this manuscript. This study was partly supported by a Research Fellowship for Young Scientists (to MA) and grants from the Japan Society for the Promotion of Science (No. 21248017 to TH) and the Ministry of Environment (No. D-0909 to TH and No. S-9-3 to MA and TH).

Ancillary