Study area
The study was conducted in the Nanhsi forest dynamics plot, with an area of 8.37 ha (Fig.
2.1). The geographic location and local climate of this plot are stated in Chapter 1 Study area. Note that the mean annual rainfall was 3,044 mm during 1998–2009, and 77% of
the precipitation occurred from May to September. The mean annual temperature was 11.5 °C (1998–2009), with a January mean of 6.7 °C and July mean of 14.9 °C.
The forest vegetation and tree data according to Yang et al. (2008) are stated in Chapter 1 Vegetation. Recall that the forest vegetation types identified in this plot included two Lauraceae-dominated forest types (the M. japonica type and the M. japonica-C. cuspidata type), one Theaceae–Fagaceae forest type (the Schima superba-C. cuspidata type), and one deciduous forest type (the Alnus formosana type).
Field survey of understory plants and tree seedlings
Analyzing assemblages of understory plants is a prerequisite for exploring the effects of neighboring communities on tree seedlings in regeneration habitats. Understory plants and tree seedling species were surveyed in 10 parallel transects with a regular spacing of 30 m in the north–south direction. Each transect was 200 m in length and 2 m in width, and it was divided into 100 quadrats of 2 m × 2 m sampling units. A total of 994 quadrats were surveyed (excluding six quadrats that were located either in water or on the bare rock of slopes) from May to July 2009 (Fig. 2.1). Understory plants (excluding tree seedlings) contained 60 families, 118 genera, and 190 species; most of them were herbs, including ferns. Tree seedlings (individuals of tree species with DBH < 1 cm) contained 19 families, 29 genera, and 36 tree species with 3,687 individuals, and the most commonly occurring families were the Lauraceae (2,605), Pentaphylacaceae (303), Adoxaceae (233), and Fagaceae (225). Both the understory plants and tree seedlings in each quadrat were measured to obtain their heights and coverage percentages.
Neighborhood communities, neighborhood factors, and environmental factors
Neighborhood communities include understory plants and trees in the upper forest strata that surround seedling assemblages in space. To explore the effects of neighboring trees, we used data from a tree census in 2006 (Yang et al. 2008), which were subdivided into trees in the shrub layer (for trees ≤ 5 m in height) and the overstory (> 5 m). We computed the basal areas (BAs) of the shrub layer (shrub BA) and overstory (overstory BA) of trees within a circle (radius = 4 m) surrounding the center of each quadrat (also see Fig. S2.1a), and we calculated tree density as the number of stems (both in the shrub layer and overstory) divided by the circle area.
Among the understory plants, tall herbs often throw dense shade on seedlings and compete with seedlings for limited resources; therefore, tall-herb (with heights exceeding 30 cm) cover and mean height in each quadrat were also recognized as neighborhood factors that might affect seedling assemblages. In summation, the neighborhood factors include overstory BA, shrub BA, tree density, and herb cover and mean height in each quadrat (Fig. S2.1a). These factors were treated by a principal component analysis (PCA), and the PCs were abbreviated as strata PCs (Fig. S2.1a). Strata PC1–4 explained 29.0%, 20.0%, 16.3%, and 13.6% of the variations, respectively. The top two factors for PC1–4 were cover and the mean height of tall herbs for PC1, shrub BA and tree density for PC2, cover and the mean height of other herbs for PC3, and overstory BA and shrub BA for PC4. Note, the neighborhood factors do not consider species identity.
Topographic factors can directly or indirectly affect species assembly (Brown et al.
2013). Topographic factors include altitude, slope, and convexity. Altitude was measured by a theodolite (Yang et al. 2008), and it was used to calculate the slope and aspect. The circular aspect angle was mapped into two orthogonal components on sine and cosine axes. The convexity of each subplot was calculated as the subplot’s altitude minus the mean altitude of eight neighboring subplots. Altitude, slope, and convexity, together with
their third degree orthogonal polynomials, were used to model the relationships between topography and community variations (Legendre et al. 2009).
Data analysis
Examination of the spatial structure of seedling assemblages in regeneration habitats Here, among the effects of neighborhood communities, we singled out the understory assemblages for the aforementioned ecological reasons. To test our hypothesis that understory plants, as the nearest neighbors of tree seedlings, are a better spatial descriptor of regeneration habitats, we needed to extract the understory plant spatial structure using the spatial patterns of understory assemblages. We purposely distinguished this factor from neighborhood factors (strata PCs) that are related to the spatial patterns of coverage (e.g., the BA if trees, or herb cover). Flowcharts for the analytical procedures are shown in Fig. S2.1.
We first derived the understory plant spatial structure from principal coordinates of neighbor matrices (PCNM) (Borcard & Legendre 1994; Borcard & Legendre 2002; Dray et al. 2006). Briefly, PCNM transforms pair-wise distances among quadrats into Moran’s eigenvector maps (MEMs) that represent spatial variables at different scales (hereinafter, PCNM variables). PCNM variables that significantly explained the spatial variation of understory assemblages were considered potential critical variables to represent the understory plant spatial structure. The significance of PCNM variables as explanatory variables (X) in redundancy analysis (RDA) for spatial variation of understory communities (Y, after being subjected to a Hellinger transformation) were determined by forward selection. We identified a minimum subset of 44 PCNM variables based on a permutation test (Jombart et al. 2009, Wagner 2013) (Fig. S2.1b, also see Fig.
S2.2a), which explained 20.1% of the variation in the understory assemblage Fig. S2.2b).
Then, a partial RDA (Fig. S2.1b) was used to partial out the covarying effects of topography (which explained 7.1% of the variation, Fig. S2.2b), and extract the pure understory plant spatial structure. Hereafter, we used uRDA1 and uRDA2 to refer to the understory plant spatial structure, and they comprise the site scores of the 44 PCNM variables ordinated on the first two RDA axes; that is, the two major axes in relation to the PCNM variables in scales, which represent the variability in understory assemblages across quadrats.
Then, we tested our hypothesis by examining whether (i) the understory plant spatial structure affected the spatial distribution of seedling assemblages, and (ii) seedling species were distributed differentially over its major gradient (using uRDA1 as a proxy).
In (i), we evaluated the relative contributions from the understory plant spatial structure, topography, and neighborhood factors (strata PCs) to explain the spatial variation of the seedling assemblages, as well as the dominant taxon groups of tree seedling species (Lauraceae, Pentaphylacaceae, Adoxaceae, and Fagaceae). The relative contributions toward explaining the variations (adjusted R2 statistic, R2a) were tested using F-statistics by forward selection in an RDA (Legendre et al. 2005; Blanchet et al. 2008; Legendre and Legendre 2012). Contributions from the neighborhood factors imply that there are neighboring effects on seedling assemblages across forest strata; notably, those spatial scales, especially those from the upper forest strata, are usually quite coarse (Metz 2012), as can be seen from the spatial autocorrelation scales of neighboring effects assessed through a spatial correlogram of Moran’s I (Borcard et al. 2011).
In (ii), we examined how seedling species were distributed over the gradient of the understory plant spatial structure and the topographic factors. Specifically, we used the first PC of the topographic factors (topographic PC1, which explained 30% of the variation in the PCA and mostly correlated to altitude, slope, and convexity) to represent
the topographic gradient. Note, as explained above, the topographic effects on the understory plant spatial structure were partialed out; that is, the abiotic and biotic gradients were orthogonal. The biplot of uRDA1 and topographic PC1 indicated the niche differentiation between understory and seedling assemblages. The confidence interval (within 0.025–0.975 quantiles) of the understory and seedling species distributed over the two axes were evaluated by bootstrapping 1,000 times.
Examining the spatial covariations between seedling species versus adults and other understory species
The neighborhood effects introduced by bio-interactions between seedlings and neighboring species (Fig. S2.1c, also see Table S2.1) were further examined by a codispersion analysis. Bio-interactions, such as dispersal limitation, positive facilitation, or competitive exclusion, can be evaluated based on positive or negative codispersion patterns in space. Codispersions were conducted for the most abundant species to assess the link between the seedling species and the top two influential tall herb species (Table S2.1), and between the seedling species and their conspecific adults. The codispersion analysis is a nonparametric technique used to assess how the presence of a pair of species covaried across a range of spatial lags (distances between points) based on cross-variograms of the pair of spatial series (Rukhin and Vallejos 2008). Briefly, the codispersion analysis we applied included three steps. First, the codispersion coefficient of two spatial series between points at a given lag was evaluated. The codispersion coefficient is defined as the cross-variograms normalized by the square root of the product of each semi-variogram. Formulas for the codispersion coefficient are given in Vallejos et al. (2015), with more statistical details in Cressie (1993) and Rukhin and Vallejos (2008). The codispersion coefficient ranges from −1 (strongest negative covariations) to
+1 (strongest positive covariations). Second, codispersion coefficients for lags and for azimuths (0–π) were mapped in a radial way onto the x-y plane (a so-called codispersion map, Vallejos et al. 2015), which facilitates detecting the directionality (anisotropy) of covariations between two spatial series (Buckley et al. 2016). Third, the significance of the codispersion coefficients was tested by comparing observed values against a spatial permutation. Specifically, we toroidal-shifted individuals’ distributions (torus permutation), which maintains the autocorrelation structure after the shifts. This torus permutation accounts for the autocorrelation in space. Significance was determined following 200 permutations using the two-tailed test, with P < 0.025.
Classifying emerging ecological groups of seedling assemblages with their specific regeneration habitats, and examining the differences of community patterns therein To assess how regeneration niches for seedling assemblages were shaped by the effects of the understory plant spatial structure, we identified different regeneration patches over the uRDA1 gradient. In practice, uRDA1 was used to cluster quadrats into spatial-structuring groups (SSGs) by the constrained clustering method (Borcard et al. 2011;
Dray et al. 2012). Five SSGs (hereafter SSG1–5) were identified along uRDA1 (plotted in Fig. 2.1; the physical and biotic attributes are listed in Table S2.2). We further examined the associations between species (both understory plants and tree seedlings) and site-groups (i.e., SSGs) by an indicator species analysis (De Cáceres et al. 2009). A generalized form of an association index, the phi-coefficient (rpb), was adopted to indicate a specific preference among species for SSGs.
To verify whether regeneration niches really differentiated among SSGs, abundance (density) and a diversity index were compared among the SSGs for different seedling stages and understory plants. Seedling stages included new recruits of seedlings (height
< 30 cm) and older seedlings (height ≥ 30 cm). For the diversity index, we used Hill’s N1 diversity index, which expresses α diversity as the effective number of species (Hill 1973;
Jost 2007). The significance of differences in abundance and the diversity index were tested by a random permutation test. Here, we did not use the torus permutation, because the SSGs were defined by uRDA1, which already accounted for the spatial autocorrelation mostly related to topography. Through this analysis, we assessed how seedling assemblages responded to the biotic effects of neighborhood understory assemblages for those regeneration patches identified along the uRDA1 gradient.
Investigating the association between seedling and adult assemblages
For each SSG, a similarity index (Jost 2007) of species composition for seedlings versus juvenile trees (those that only reached shrub layer and whose DBH was less than 5 cm) and adults (trees other than juvenile trees) in each forest type was determined. Here, juvenile trees were also incorporated into the analysis to clarify the transition in association across life-history stages to better understand the coupling (or uncoupling) between seedling/adult stages. This analysis clarified how regeneration niches affect subsequent community assembly.
Computation
All the data analyses were performed in R (R Development Core Team 2016), with our data and codes available at https://github.com/cywhale/spatially_structured_regen. Here, PCNM, forward selection, and RDA were performed using the R packages “PCNM”,
“packfor”, and “vegan” (Legendre et al. 2012; Dray 2016; Oksanen et al. 2016), respectively. The spatial correlogram of Moran’s I was evaluated using the “spdep”
package (Bivand et al. 2013; Bivand and Piras 2015). Codispersion patterns were tested by the torus permutation with the “geoR” package (Ribeiro Jr and Diggle 2016). An
indicator species analysis was performed using the “indicspecies” package (De Cáceres et al. 2009). A similarity index was computed using the “vegetarian” package (Charney and Record 2012).
Results
Understory plant spatial structure as the spatial descriptor of spatially structured regeneration habitats
The understory plant spatial structure that ordinated on the first RDA axis (uRDA1) was the most influential factor in terms of explaining the variation in seedling assemblages, more so than the topographic factors (Table 2.1). We noted that uRDA1 negatively correlated with tall herb cover (Fig. 2.2a) and positively correlated with understory and seedling species diversity, and seedling density (Fig. 2.2b–d). Moreover, tall herbs and uRDA1 had a similar fine-scale autocorrelation structure (finer than 25 m; Fig. S2.3a, b), suggesting that tall herb species played important roles in characterizing the understory plant spatial structure.
In addition to the understory plant spatial structure and topographic factors, neighborhood factors also contributed significantly to the seedling assemblages (strata PCs in Table 2.1). Nevertheless, we found that the dominant influential factors varied among taxonomic assemblages. While the Lauraceae and Fagaceae were mostly affected by uRDA1, the Pentaphylacaceae and Adoxaceae were mainly affected by the topographic slope and altitude, respectively (Table 2.1). More specifically, the Adoxaceae occupied two extremes: Viburnum luzonicum at higher elevations (ridges) and Viburnum taitoense at lower elevations (also see Fig. 2.2e). Eurya species (Pentaphylacaceae) generally inhabited steeper slopes. The high-order polynomials of topographic factors
hardly contributed, suggesting that the variation in seedling assemblages was only related linearly to the topographic factors.
The distribution of seedling species over the gradient of uRDA1 and topographic PC1 (Fig. 2.2e) revealed that their association with regeneration habitats differed because of the heterogeneity of the understory assemblage or topography. Species clusters could be distinguished along the uRDA1 axis, accompanied by a co-occurrence of seedling species and tall herb species. For example, Fagaceae seedlings and Dryopteris formosana were at the positive end of uRDA1, while Lauraceae seedlings and Alpinia shimadae were near the zero value of uRDA1, and Eurya loquaiana (Pentaphylacaceae) was at the negative end of uRDA1, with the tall herb Miscanthus sinensis at the distal end. A spatial interspecific correspondence was also revealed by showing how these tall herb species contributed to the variation in seedling assemblages (Table S2.1). Adult trees also exhibited a similar correspondence, but they were not as influential as tall herb species, except for the Adoxaceae (Table S2.1).
In summation, our hypothesis that the understory plant spatial structure is a better surrogate as a spatial descriptor of regeneration habitats than topographic structuring is supported by these results, because (i) uRDA1 explained most of the seedling distribution and (ii) seedling species were distributed differentially over uRDA1. Therefore, we can use uRDA1 to classify emerging ecological groups of seedling assemblages with their specific regeneration habitats.
Codispersion analyses showing how seedlings covaried with adults and understory neighbors
Codispersion patterns between seedling and adults varied among taxonomic groups. We found that the codispersion coefficients of C. cuspidata (Fagaceae) and Eurya leptophylla
(Pentaphylacaceae) were significant relative to the torus permutation in all directions (Fig.
2.3c, d). The patchy-distributed seedlings of Fagaceae and Pentaphylacaceae suggest a dispersal limitation from adult trees. In contrast, Litsea acuminata (Lauraceae) had widespread seedlings, but it seldom exhibited significant codispersion coefficients relative to the torus permutation (Fig. 2.3a). The codispersion coefficients of Machilus spp. (Lauraceae) were rarely significant relative to the torus permutation in most y-direction lags (Fig. 2.3b), whereas those of V. taitoense (Adoxaceae) were mostly significant, except for some clumped lags (within 30 m) (Fig. 2.3e).
Codispersion analyses between seedlings and understory tall herb species indicated that they could be co-occurring or segregating. The segregation patterns, as shown in the two most widespread Lauraceae seedlings versus M. sinensis (Fig. 2.3f, g), were more significant in the −x (left) through +y (upper) direction, which was inhabited by M.
sinensis. The codispersion coefficient of V. taitoense was not significant relative to the torus permutation when it was evaluated with two covarying tall herbs (Fig. 2.3j, o). This is consistent with the result showing that tall herbs were much less influential factors than topography (altitude) for Adoxaceae seedlings (Table 2.1). Both E. leptophylla and C.
cuspidata co-occurred with tall herb species. Eurya leptophylla significantly co-occurred with Arachniodes rhomboidea in most +x (right) directions (Fig. 2.3i), and with D.
formosana, especially in some +x (right) through +y (upper) directions (Fig. 2.3n).
Castanopsis cuspidata co-occurred with both D. formosana and Carex spp. (Fig. 2.3h, m) and the codispersions were significant in almost all directions. Seedling species were not only distributed differentially over uRDA1 (as mentioned above), but they also formed close spatial relationships with understory species (i.e., co-occurrence or segregation according to the codispersion test).
Differences in species diversity and seedling density reveal that seedling assemblages responded differently among the SSGs
Recall that the five SSGs were identified using constrained clustering along uRDA1.
Different regeneration patches can be further recognized by (i) differences in community patterns (diversity and abundance of seedlings and understory plants) among the SSGs, and (ii) specialization in species–SSG relationships (as revealed by the indicator species analysis, Table S2.3).
In (i), for understory species, each SSG had significantly lower diversity than expected by random chance (Fig. 2.4c). The lower-than-expected understory species diversity in each SSG was due to the clumped tall herb species, which covered larger areas of neighboring quadrats than other understory species. Understory plants in SSG4 were overwhelmingly dominated by M. sinensis (with a very high phi-coefficient of rpb = 0.82 in the indicator species analysis, Table S2.3), resulting in much lower diversity.
For older seedlings (height > 30 cm), SSG1–4 had significantly lower diversity than expected by random chance (Fig. 2.4b). In contrast, the species diversity of new recruits (height < 30 cm) was usually not lower than that expected by random chance (Fig. 2.4b, in SSG1, 3–5). The difference between older seedlings and new recruits may be explained by the fact that older seedlings were continuously shaped by neighboring tall herbs, while new recruits were affected to a lesser degree.
SSG5 had the greatest seedling density among the SSGs and a significantly greater density of older seedlings than expected by random chance (Fig. 2.4a). The seedling density of SSG4 for both older seedlings and new recruits was the lowest among the SSGs and lower than that expected by random chance. SSG3, which is in the neighborhood of SSG4, had a higher density of older seedlings but a lower density of new recruits than expected by random chance. Seedling densities showed clearer differences among the
regeneration patches of SSG3–5. In contrast, SSG1 and SSG2 exhibited random patterns of moderate values and smaller variations of confidence interval.
In (ii), SSG1 and SSG2 were the widespread regeneration habitats in valleys and slopes that could accommodate different sets of seedling species (Table S2). However, no indicator species of tree seedlings was dedicated exclusively to SSG1 or SSG2 (Table S2.3, only V. taitoense was associated with the union of SSG1 and SSG2). Therefore, we do not regard SSG1 and SSG2 as patches; instead, they represent the baseline with which to compare the community patterns in the other regeneration patches (SSG3–5) below.
SSG4 had the lowest seedling density and species diversity, with one indicator species of seedlings, Eurya chinensis (rpb = 0.16, Table S2.3). However, different indicator species of the seedlings Acer kawakamii (rpb = 0.32, Table S2.3) and E. leptophylla (rpb
= 0.26) were found in SSG3. SSG5 maintained a very specific community pattern with the most diverse and abundant older seedlings. Many indicator species of seedlings were specific to this patch, including C. cuspidata, S. superba, and others (detailed in Table S2.3).
Association between seedling and adult assemblages varies among different SSGs In SSG1 and SSG2, we found a high similarity index between seedling assemblages versus juvenile tree and adult assemblages (approximately 0.72–0.9, Table S2.2); i.e., the community assemblies across life-history stages were coupled. In SSG4, uncoupling was substantial, as can be seen in its low similarity (0.39–0.4) of species composition between seedlings versus adults in Lauraceae-dominated or deciduous forest types (Table S2.2).
In SSG3, a discrepancy between seedling and adults was also found (a relatively low similarity index of 0.62, Table S2.2), but its similarity index was still higher than that of SSG4. SSG5, which is located in the Theaceae (S. superba)–Fagaceae (C. cuspidata)
forest type, seedling assemblages exhibited high similarity with juvenile and adult trees (Table S2.2), suggesting that the older seedlings have a higher chance of surviving into the next life-history stage. Together with the results of the most diverse and abundant older seedlings found here, SSG5 was an optimal habitat for species assemblages to persist across life-history stages.
forest type, seedling assemblages exhibited high similarity with juvenile and adult trees (Table S2.2), suggesting that the older seedlings have a higher chance of surviving into the next life-history stage. Together with the results of the most diverse and abundant older seedlings found here, SSG5 was an optimal habitat for species assemblages to persist across life-history stages.