What drives phylogenetic and trait clustering on islands?

Environmental filtering is an important assembly process that structures plant communities and is commonly inferred from taxonomic, functional trait and phylogenetic patterns. However, while these approaches can be informative, the influence of other co-occurring processes on community diversity, such as competitive exclusion, remains poorly understood. By combing functional traits and a phylogeny of woody plants across anthropogenically created islands, we aim to explore the ways in which environmental filtering and competitive exclusion can simultaneously influence community assembly processes. We expect that communities on smaller islands, where competition for limited space and resources is more intense, should be functionally and phylogenetically less clustered than those on larger islands because this more intense competition should reduce the coexistence of the most closely related or functionally similar species. We used ten functional traits and a phylogeny of 76 woody plant species to assess species diversity and similarity within communities across an island area gradient. We combed functional traits and phylogeny into a functional–phylogenetic distance matrix and calculated the communities’ mean functional-phylogenetic distance (MFPD) and its standardized effect size (SES.MFPD) as measures of ecological similarity. As expected, species were more phylo-functionally similar to one another than expected by chance within islands and this underdispersion grew stronger with island area, indicating that while islands generally contained clustered communities, environmental filtering and competitive exclusion were both likely occurring. By integrating species abundance distributions with community similarity, we found that the most abundant species were phylo-functionally similar to the least abundant species. Species richness increased with island area, as expected, but the additional species found only on larger islands tended to have low abundances, providing opportunities for rare species to persist. With environmental filtering narrowing the number of species that can persist, the loss of phylo-functionally closely related rare species on smaller islands was likely caused by competition or stochastic removals, leading to greater species dissimilarity than on larger islands. On larger islands, the clustered patterns are likely to be the result of a combination of competitive exclusion caused by resource limitation and environmental filtering.

islands, where competition for limited space and resources is more intense, should be functionally and phylogenetically less clustered than those on larger islands because this more intense competition should reduce the coexistence of the most closely related or functionally similar species. Methods We used ten functional traits and a phylogeny of 76 woody plant species to assess species diversity and similarity within communities across an island area gradient. We combed functional traits and phylogeny into a functional-phylogenetic distance matrix and calculated the communities' mean functional-phylogenetic distance (MFPD) and its standardized effect size (SES.MFPD) as measures of ecological similarity. Results As expected, species were more phylofunctionally similar to one another than expected by chance within islands and this underdispersion grew stronger with island area, indicating that while islands generally contained clustered communities, environmental filtering and competitive exclusion were both likely occurring. By integrating species abundance distributions with community similarity, we found that the most abundant species were phylo-functionally similar to the least abundant species. Species richness increased with island area, as expected, but the additional species found only on larger islands tended to have low abundances, providing opportunities for rare species to persist. Conclusions With environmental filtering narrowing the number of species that can persist, the loss

Introduction
Biodiversity on islands has been the focus of much research over the past few decades, after the Theory of Island Biogeography (MacArthur and Wilson 1967) was proposed. Since community assembly processes are often non-random, environmental filtering, i.e., an abiotic filter directly acting on species establishment, persistence and competitive ability (Kraft et al. 2015;Cadotte and Tucker 2017), inferred from taxonomic, functional trait and phylogenetic patterns, has been one of the prevailing hypotheses of selective extinction on islands. However, even with these approaches, the effect of other processes on island assemblages, such as competitive exclusion, influenced by species niche and competitive differences (Mayfield and Levine 2010), especially how it might contribute to both community under-or over-dispersion, still remains poorly understood.
In addition to stochastic processes (Chase 2007), environmental filtering and competitive exclusion are believed to shape community diversity patterns and can be reflected in the patterns of species trait or phylogenetic similarities. Environmental filtering is commonly inferred as the process driving functional or phylogenetically clustering on island biotas, examined by negative standardized effect size (SES) values of functional or/and phylogenetic diversity indices (May et al. 2013;Riemann et al. 2017;Si et al. 2017;Cravens et al. 2018), though it is also hypothesized that competition can lead to clustering (Mayfield and Levine 2010;Cadotte and Tucker 2017).
Combing the Theory of Island Biogeography (MacArthur and Wilson 1967) with functional and phylogenetic diversity (e.g., Si et al. 2017), we predicted that plant community ecological similarity will be affected by island area. Specifically, if environmental filtering is dominant, the negative SESs of phylogenetic and functional diversity indices would decrease with decreasing island area, with a stronger clustering pattern and high habitat filtering on smaller islands (Fig. 1d). With environmental filtering, we should expect that its signature is stronger on smaller islands because they should be relatively homogeneous while large islands are more likely to contain greater habitat heterogeneity (Hortal et al. 2009;Allouche et al. 2012;Si et al. 2017). When competitive exclusion was taken into account, it would depend on the relative strength of species niche differences and competitive ability differences (i.e., relative fitness differences) (Mayfield and Levine 2010;HilleRisLambers et al. 2012). If there's no environmental filtering, and if species niche differences and competitive ability differences are both unimportant for coexistence, then the community should show random patterns (Fig. 1c). If species niche differences determine coexistence, competitive exclusion might favor an overdispersion pattern, and the positive SESs of phylogenetic and functional diversity indices decrease with island area (Fig. 1a) because of spatial partitioning, habitat heterogeneity and mass effects will be more important on larger islands. Whereas, when species' competitive ability differences play a more important role, competitive exclusion could lead to a clustered community pattern, with the negative SESs of phylogenetic and functional diversity indices decreasing with island area (Fig. 1b), due to lower stochasticity and the fact that groups of high fitness species should reach higher abundances on larger islands. However, given that competitive interactions do not operate in isolation from filtering, community assembly processes on smaller islands should be influenced more by environmental filtering than on larger islands, since most species fail to survive on smaller islands because of the extreme environmental conditions, such as increased disturbance effects on smaller islands (Ewers and Didham 2006;Haddad et al. 2015;Zhang et al. 2021). As a consequence, if environmental filtering and competitive exclusion drive community assembly processes simultaneously (Fig. 1e), even less intense competitive exclusion should result in extinctions or prevent colonizations on small islands, and thus resulting in coexisting species on smaller islands being more dissimilar than on larger islands (MacArthur 1972).
However, it's difficult to distinguish between environmental filtering and competitive exclusion, when they were both caused by resource limitation (abiotic factors, such as insufficient amounts of certain soil elements) on larger islands. Here then, differences in species competitive ability should be more important than niche differences (Fig. 1b), resulting in the niches of rare species being occupied by ecologically similar common species, and rare species facing an elevated risk of competitive exclusion (Umaña et al. 2017), and so we expect a weakened overdispersion pattern on larger islands where rare species should persist better than on the more competitive smaller islands.
Here, in a land-bridge island system, the Thousand Island Lake region, China, we integrated functional traits and the phylogeny of woody plants into measures of ecological similarity, to explore the importance of assembly mechanisms in structuring island communities. We expected one of two potential scenarios: (1) for island communities to show clustering caused by environmental filtering and competitive exclusion, and with community similarity increasing with island area (i.e., common species are phylogenetically and functionally similar to rare species).
(2) Alternatively, an observed overdispersion pattern resulting from competitive exclusion (i.e., common species are phylogenetically and functionally different to rare species) with community similarity decreasing as island area increases.

Study site and field survey
Formed by dam construction in 1959, Thousand Island Lake (TIL) is a large-scale anthropogenic hydroelectric reservoir located in Zhejiang Province, eastern China (29°22′-29°50′N and 118°34′-119°15′E). TIL has a total water surface area of ca. 540 km 2 and has 1078 land-bridge islands with an area ranging from 0.25 ha to 1154 ha when the maximum water level (108 m a.s.l.) was reached (Yu et al. 2012). Before the dam construction (1959), forests on the islands were clear-cut (Liu Fig . 1 A hypothesis of island area on plant community structure (Mayfield and Levine 2010;HilleR-isLambers et al. 2012;Si et al. 2022). SES.MFPD measured the community structure's deviation from null random species assemblages, i.e., the clustering pattern (SES.MFPD < 0) and overdispersion pattern (SES.MFPD > 0). We expected that if environmental filtering is dominant, the negative SEE.MFPD would increase with island area (i.e., the ecological similarity decreases with island area). When competitive exclusion was taken into account, the ecological similarity would increase with island area (no matter SES.MFPD is positive or negative) et al. 2019), and the major vegetation type is now secondary successional forest dominated by Masson pine (Pinus massoniana) without alien woody species. The climate is subtropical and influenced by the monsoon (i.e., hot-wet summers and colddry winters), with a mean annual precipitation of 1430 mm and a mean annual temperature of 17.0℃ (Liu et al. 2019). During 2009-2010, we established forest dynamic plots on 29 islands in TIL (Table S1 and Fig. 2). On smaller islands (≤ 1 ha), we set up contiguous 5 m × 5 m subplots spreading across the entire island ( Figure S1, Liu et al. 2018). On larger islands (> 1 ha), we constructed two or three transects (consisting of 5 m × 5 m subplots) on each island to cover the edge-interior contrasts of an island ( Figure S1, Liu et al. 2018). The number of subplots per island ranged from 7 on the smallest island to 598 on the largest island (5082 subplots across all islands, Table S1, Zhang et al. 2021). Sampling effects caused by variation in number of subplots were controlled in the data analyses (see below). We recorded the identity of all woody plant individuals with DBH greater than 1 cm in the plots. A total of 76 woody plant species were recorded on the 29 study islands.

Functional traits and community phylogeny
Ten functional traits of all woody plant species were measured (Table S2): leaf area, leaf thickness, leaf dry matter content, specific leaf area, leaf chlorophyll content, leaf stomatal density, leaf C:N ratio, leaf N:P ratio, wood density and seed mass. See the measurement details in Zhang et al. (2021). These traits were associated with fundamental aspects of species life history: dispersal, establishment and persistence (Zambrano et al. 2019). All the measured traits were not highly correlated (Pearson r < 0.600, Table S3). We quantified phylogenetic trait conservatism using Blomberg's K (Blomberg et al. 2003) and Pagel's Lambda (Pagel 1999), and most traits had strong and statistically significant phylogenetic signals (Table S4).
We built our phylogenetic tree with the package 'V.PhyloMaker'  in R 3.5.1 (R core Team 2018), which we used for further analyses ( Figure S2). All 76 woody plant species in our study were included in the mega-tree implemented in 'V.PhyloMaker', which includes 74,533 species and all families of extant vascular plants. Fig. 2 Map of the islands used in this study in the Thousand Island Lake (TIL). The 29 study islands are shown in red (Liu et al. 2020) Vol.: (0123456789) Functional-phylogenetic distance matrix To overcome the shortcomings of approaches that are only based on functional traits or phylogenies alone, Cadotte et al. (2013) integrates the information provided by functional traits and phylogeny into a functional-phylogenetic distance matrix. When combing functional traits and phylogeny, we used the function comparative.comm in the R package 'pez' (Pearse et al. 2015). The comparative.comm function creates a community comparative ecology object, by combining the information of phylogeny of species (phylogenetic tree), community matrix (with species as columns and rows as communities) and data frame of species traits. With this community comparative ecology object, we made a co-existence matrix (functional-phylogenetic distance matrix) based on functional traits or phylogeny by a weighting parameter (a), using the function funct.phylo.dist in R package 'pez' (Pearse et al. 2015). When a = 0, the distance matrix only captures functional distances; when a = 1, the distance matrix reflects phylogenetic distances alone. At intermediate values of a, both functional and phylogenetic distances contribute to the distance measures, and the optimal a can be selected according to the highest adjusted R 2 of the final model (see below).

Ecological similarity within and between islands
To estimate the ecological similarity of plant communities (community similarity) within islands, we calculated the communities' mean pairwise functionalphylogenetic distance (MFPD pw ) and its standardized effect size (SES.MFPD pw ), which quantifies the average functional-phylogenetic relatedness between all possible pairs of species in an assemblage (Cadotte et al. 2013;Qian et al. 2019); as well as the mean nearest functional-phylogenetic distance (MFPD nn ) and its standardized effect size (SES.MFPD nn ), which quantifies the average functional-phylogenetic relatedness of every species to its most closely related species in the phylogeny in an assemblage (Dehling et al. 2014). MFPD pw and MFPD nn were calculated by the function.ses.mpd and.ses.mntd in the R package 'pez' (Pearse et al. 2015), and their standardized effect size values (SES.MFPD pw and SES.MFPD nn ) were calculated by keeping constant species richness in each community and species shared between communities (Kembel et al. 2010).
SES.MFPD was calculated as SES.MFPD = (M FPD observed − MFPD randomized )/(sdMFPD randomized ), where MFPD observed was the observed MFPD, and MFPD randomized and sdMNTD randomized were the expected mean and standard deviation of the randomized assemblages. To control the potential sampling effect (number of plots and individuals), we used a null model as follows. After including all 76 species into a pool, we simulated randomized assemblages (null communities) based on a random draw (with replacement, repeated 1000 times) of the same species richness as observed on each of our 29 sampled islands. When SES.MFPD values were zero, the species were distributed randomly. When SES.MFPD values were negative, the observed SES.MFPDs were less than that of the randomized assemblages, meaning that species were more closely phylo-functionally related than that would be in a randomized assemblage, i.e., species were phylo-functionally clustered (Webb et al. 2002). Inversely, when SES.MFPD values were positive, the observed SES.MFPDs were greater than that of the randomized assemblages, meaning that species were more distantly phylo-functionally related than would be in a randomized assemblage, i.e., species were phylo-functionally overdispersed (Webb et al. 2002).
In a review of the effects of fragmentation in the TIL system, Wilson et al. (2016) relayed that isolation (distance to mainland) had negligible effects on plant species richness. Further, in this study, species richness, phylo-functionally richness, phylo-functionally diversity, MFPD and SES.MFPD on islands also failed to exhibit a significant relationship (P > 0.05) with isolation (distance to mainland and distance to nearest island), tested by simple linear regression. Thus, we focused on island area in the present study. Simple linear regression models were used to assess whether the MFPD and SES.MFPD values shift with island area (log-transformed). By changing a value which ranged from 0 to 1 (in steps of 0.05), we calculated the adjusted R 2 of each regression model. Since SES. MFPD is the standardized effect size of MFPD, which controls for the number of species observed in samples, we selected a value when the highest adjusted R 2 was reached for SES.MFPD. We calculated both nonweighted and abundance-weighted values during the a optimization process ( Figure S3). And we found that, only with the non-weighted dataset, MFPD nn and SES. MFPD nn always had a significant relationship with island area (the highest adjusted R 2 (0.481) was reached when a = 0.80 for SES.MFPD nn ) Therefore, we used a = 0.80 for the following data analysis, and this value corresponds to a greater contribution from the phylogeny than the functional traits.
The community similarity between islands was calculated using the R package 'picante' (Kembel et al. 2010). The function comdist was used to calculate MFPD pw and the function comdistnt calculated MFPD nn , with non-weighted and abundance weighted dataset respectively. We used the Principal Co-ordinates Analysis (PCoA) with the R package 'FD' (Laliberté et al. 2014) to visualize the community similarity between islands. When we calculated the functional-phylogenetic distance matrix at a = 0.80, MFPD pw was weakly represented by PCoA (the first five PCoA axes explained < 35% of variance, the first two axes explained < 20% of variance, Table S5), indicating that the detected patterns of community similarity between islands were mostly random (Silk et al. 2018). However, the first five PCoA axes for MFPD nn explained 99% of variance (Table S5). Thus, we only used the mean nearest functional-phylogenetic distance (MFPD nn ) for all the analysis.
To quantify community diversity, we also calculated the functional-phylogenetic richness and functional-phylogenetic diversity, using the functional-phylogenetic distance matrix. As an equivalent to functional richness, functional-phylogenetic richness was calculated by the function dbFD in R package 'FD' (Laliberté et al. 2014), which didn't include abundance in the calculation. Similar to phylogenetic diversity, functional-phylogenetic diversity was calculated by the function.pd in R package 'pez' (Pearse et al. 2015), which also didn't include abundance in the calculation. Therefore, we only calculated the non-weighted functional-phylogenetic richness and diversity. Both of these measures are technically richness measures since they involve summations of distances (e.g., Tucker et al. 2017), and so we expect them to be correlated.

Integrating species abundance distributions and SES. MFPD
To verify that environmental filtering and competitive exclusion both drive community assembly process on islands in TIL region, we quantified species ecological similarity (i.e. species functional-phylogenetic distance). To quantify the contribution of each species to community MFPD, we integrated the standardized effect size for SES.MFPD with species rank abundance distribution, by following the method proposed by Mi et al. (2012). In the first step, we calculated the SES.MFPD for the most and the second most abundant species in the sample, through the null community generated by randomizing two names of species 1000 times across the functional-phylogenetic matrix. Next, we added the third abundant species into the sample and calculated SES.MFPD. This process was repeated by adding increasingly less abundant species into the sample.
Trends in SES.MFPD (as well as MFPD) along the species rank abundance axis can reveal the contribution of an individual species to the overall community MFPD. The significances of trends were tested using the Mann-Kendall test with the R package 'Kendall' (McLeod 2011). We also tested for the segmented regression in trends using a segmented regression model with the R package 'segmented' (Muggeo 2017). If there is a decreasing trend in SES.MFPD (as well as a decreasing trend in MFPD) along the species rank abundance axis (from most to least abundant), it indicates that the new added species (less abundant) were more ecologically similar to already added species (more abundant) than randomization (null expectation). In other words, the new added species contributed less to the overall community MFPD. On the contrary, if there was an increasing trend in SES.MFPD (as well as an increasing trend in MFPD) along the species rank abundance axis, the new added species was more ecologically dissimilar to already added species, and contributed more to the overall community MFPD.

Results
Species richness significantly increased with island area (adjR 2 = 0.70, P < 0.001), with 53 species occurring on the largest island and 8 species on the smallest island. The total abundance per species across the 29 islands ranged from one observed individual to over 74,000 individuals, with 20 species having fewer than 10 individuals. Community functionalphylogenetic richness and functional-phylogenetic diversity significantly increased with island area as well (Fig. 3).
Community MFPD nn and SES.MFPD nn values significantly decreased with increasing island area (Fig. 4abc, Table S6), indicating that the species within island plant communities were increasingly similar with one another with increasing island area. Meanwhile, SES.MFPD nn values on most islands (23 out of 29 islands) were below zero, especially for large islands (Fig. 4c; corresponding to the scenario in Fig. 1be). Besides the combined functional-phylogenetic distance, these results were also supported by phylogenetic and functional distance, respectively ( Figure S4, S5 and S6).
When including less abundant species into the analysis, MFPD nn and SES.MFPD nn both showed a significant decreasing trend with species abundance rank (MFPD nn : Mann-Kendall trend test tau statistic = -0.868, P < 0.001; SES.MFPD nn : Mann-Kendall trend test tau statistic = -0.330, P < 0.001; Fig. 5), Fig. 3 The functionalphylogenetic richness (a = 0.80, non-weighted) and functional-phylogenetic diversity increase with log-transformed island area, using the sample linear regression. The size of the grey symbol scales with species richness Fig. 4 The mean nearest functional-phylogenetic distance (MFPD nn ) (a, b) and its standardized effect size (SES.MFPD nn ) (c, d) decrease with log-transformed island area. The size of the grey symbol scales with species richness. The dashed horizontal line in c and d denotes the null expectation indicated that the less abundant species was more ecologically similar to more abundant (already included species) than expected. Our analyses detected a breakpoint separating the 64 least abundant species (abundance < 2000) and the 12-most abundant species (abundance ≥ 2000), with a more rapid decrease for the abundant species (Fig. 5).
Here, we also found an interesting species distribution pattern that species occurring on small islands were generally a subset of the species that occurred on large islands ( Figure S7). Species richness increased with island area (Table S1), and the additional species found only on large islands were the least abundant species (i.e., rare species, Figure S7) which were phylo-functionally similar to the more abundant species.

Discussion
Combining functional traits and phylogenies, we found that communities on large islands were more phylo-functionally clustered than on small islands, which could be ascribed either to scenario Fig. 1b or 1e. Moreover, rare species were phylo-functionally similar to abundant species. Since environmental filtering is important for community assembly on islands, we could assume that environmental filtering and competitive exclusion likely simultaneously drove the clustered pattern on islands (Fig. 1e).
The species distribution pattern underpins the observation that plant communities on larger islands were more phylo-functionally clustered than on smaller islands. The loss of similar and closely related rare species on smaller islands is expected under competitive exclusion with similar but better competing species. Here, we assume that most species fail to succeed on small islands because of environmental filtering, i.e., harsher disturbance (Zhang et al. 2021). Thus, with environmental filtering narrowing the range of potential species, competitive exclusion further eliminated similar rare species on smaller islands (Fig. 1e). Hu et al. (2016) showed that environmental filtering played a significant role in plant community assembly processes, especially during the seedlingsto-saplings life-stage transition in our study system (the TIL region). Si et al. (2017) also highlighted the importance of environmental filtering on bird communities in the TIL region. Not limited to our study system, studies have also reported evidence for the important role of environmental filtering in other systems such as in fragmented forest patches (Sonnier et al. 2014), mountain grasslands (Dainese et al. 2015) and coastal dune habitats (Rolo et al. 2018).
In general, habitat heterogeneity increases with area, creating additional niches and allowing for increased species diversity in large fragments (Gardner and Engelhardt, Yu et al. 2012). Liu et al. (2018) conducted a principal component analysis with nine environment variables on the 29 study islands studied Fig. 5 The mean nearest functional-phylogenetic distance (MFPD nn ) (a, b) and its standardized effect size (SES.MFPD nn ) (a = 0.80, non-weighted) with species abundance rank distribution. The different color of dots represents different species abundance groups. The dashed vertical lines and light grey shade areas represent breakpoints and its 95% confidence intervals. The dotted lines represent the fitted curve of segmented linear regressions, with a significant Mann-Kendall trend here (e.g., total soil nitrate, total soil phosphorus, relative potential moisture of the soil, and litter layer depth) to visualize the difference in environmental variables between islands (i.e., habitat heterogeneity), and found that habitat heterogeneity increased faster with area among smaller islands than among larger islands. Working on islands in the same system, Si et al. (2017) also showed that additional niches were available on larger islands in the TIL region. Thus, we can assume that an individual smaller island has less habitat heterogeneity than on larger islands.
With a higher ecological similarity of plant communities within and among large islands ( Figure S8, Liu et al. 2018), where the least abundant species were similar to more abundant species and with one another (Fig. 5), indicating that environmental filtering and competitive exclusion both might select similar species. Even though competitive exclusion is less likely over a larger area, because there is more spatial and spatial-temporal heterogeneity on larger islands, the competitive exclusion caused by resource limitation on larger islands is intimately tied to, and impossible to distinguish from environmental filtering (abiotic factor, such as lack of soil nutrients) (e.g., Kraft et al. 2015;Cadotte and Tucker 2017).
For instance, our island system is phosphorous limited (leaf N: P ratio > 16), and soil available phosphorous content in adjacent mainland were 1.5 times higher than soils on islands (Zhong et al. 2022), and so plants that can maintain growth and reproduction under lower P concentration might be the superior competitors (Tilman 1982). This means that they could outcompete dissimilar, more P-limited, species (Mayfield and Levine 2010). Reflecting this limitation, leaf P concentration decreased with island area (adjR 2 = 0.21, P < 0.001). Competitive exclusion might eliminate all but species with a low P tissue concentration, leaving species that have similar P concentrations, and thus driving the clustering pattern (Mayfield and Levine 2010). However, this elimination of high P-requiring species could be interpreted as environmental filtering, but in reality, is the combination of both environmental conditions and competition under those conditions .
In addition, compared to SES.MFPD pw , SES. MFPD nn had a strong explanatory power for the variance between islands (Table S5, Figure S8). This is not surprising, since this is commonly observed for angiosperms (Silk et al. 2018;Massante and Gerhold 2020), and which the majority of our species were. SES.MFPD pw quantifies the overall clustering of taxa on a functional-phylogenetic tree (i.e., structure deeper in the tree), while SES.MFPD nn quantifies the extent of terminal clustering (Webb et al. 2002;Qian et al. 2019). This indicated that here competition happens between close relatives but not deep in phylogeny.
In the TIL system, we found that the most abundant species tended to be ecologically similar to the least abundant species, a different pattern from some mainland forests. For instance, in the evergreen shrubs in the California chaparral, plant communities that were functionally clustered (convergence) but phylogenetically overdispersed driven by the Mediterranean-type environments (Ackerly 2004). Meanwhile, Mi et al. (2012) found the evidence that common species were phylogenetically dissimilar to rare species in six of the nine gap-dominated forests, and Umaña et al. (2017) found that rare species tended to be functionally dissimilar to common species in the six out of eight forests (including temperate forests, pine mixed forest, tropical forests and moist forest). These studies suggested that community assembly is a very complicated process, and could be simultaneously driven by the interplay of niche and neutral processes. Further, patterns uncovered in mainland systems might not transfer well to island systems where environmental limitation should be more important.
During the dam construction in TIL, forests were clear-cut on islands (before1959), and the major vegetation type is now secondary successional forest (Liu et al. 2019). After 50 year's regeneration, this disturbance-dominated secondary forest is quite young. The forests of Mi et al. (2012) and Umaña et al. (2017) have survived for at least 100 years. As Gaston (2012) expressed, we can not assume that the same ecological processes apply to all assemblages, and so the regenerating forest in TIL might be subject to unique paradigms on habitat fragmentation, which is different from the old-growth forest (Liu et al. 2019).

Conclusion
This study indicated that in an island system, community assembly processes might involve both environmental filtering and competitive exclusion. On smaller islands, even though most species failed to maintain populations here, likely due to environmental filtering (extreme environment conditions), competitive exclusion might have also removed closely related rare species. On larger islands, the clustered patterns are likely to be the result of a combination of competitive exclusion caused by resource limitation and from environmental filtering. Under these selective extinction processes, the most abundant species were ecologically similar to the least abundant species, indicating that the least abundant species (rare species) might be locally rare because their niche has been occupied by ecologically similar species (most abundant species).