Study Area

The study site is located in the Guilin karst mountainous area (18°53′–19°20′N, 108°58′–109°53′E) in the Guangxi Zhuang Autonomous Region, China. Karst topography is a special landscape characterized by the dissolution of one or more layers of soluble bedrock in soil substrate. The soil, mainly composed of black and yellow lime soil, has a thin soil layer and exposed carbonate rocks at the floor surface of mountain areas. The climate of this area is sub-tropical with monsoons and abundant rainfall concentrated in the growing season. The annual average temperature is 19°C with the warmest days occurring in August and the coldest days occuring in January when the average temperature is 8°C followed by a frost-free period of 309 days (the China Meteorological Data Service Center;  http://data.cma.cn). Due to these heterogeneous conditions, tree species richness in this area is limited. The representative dominant species are Cyclobalanopsis glauca, Alchornea trewioides, Alchornea davidii, Pittosporum planilobum, Radermachera sinica, Cyclobalanopsis glauca, Fordia cauliflora, Celtis sinensis, Mallotus philippensis, Elaeagnus pungens.

Field Sampling

Field surveys were carried out from July to September 2017 and were done randomly within the woody plant Cyclobalanopsis glauca dominated areas. We selected four sampling locations: the BeiTou village site (BT, 24°44ˊ42.93"N,110°25ˊ19.87"E), the ZhangJia village site (ZJ, 25°18ˊ19.43"N,110°15ˊ49.34"E) the YanPo mountainou site (YP, 25°19ˊ10.08"N,110°15ˊ19.06"E), and the BaiSha town site (BS, 24°49ˊ52.99"N,110°22ˊ43.74"E). A total of twenty 20 m × 20 m plots were established in BT (7 plots), ZJ (3 plots), YP (6 plots) and BS (4 plots). In this study, we included all of the woody plants with a diameter at breast height (dbh) ≥1 cm and the height of each individual plant was measured from the base to the highest point within each plot. We also obtained information on altitude, topographic position and field observations at a scale of 20 m × 20 m using the JRBP Geographic Information System (GIS). The nomenclature of species follows Flora of China (English edition;  http://www.efloras.org) and expert identification. Site environmental conditions and species richness in BT, ZJ, YP and BS are listed in Table 1.

Table 1.

Site Conditions and Species Richness in 20 Plots for Cyclobalanopsis glauca Forest of the Guilin Karst Mountainous Areas, China.

Measurement of Environmental Factors

At each 20 m × 20 m plot location, soil samples were collected from five points, including each of the four corners and the center of each plot. A core of the top 20 cm of soil was taken at each point. Soil samples were air-dried, then sieved (2 mm) to remove bulky organic material. Soil water content (SWC, %) was calculated by subtracting the dry weight from the water saturated weight of each sample. After taking this moisture value, the five samples were thoroughly mixed prior to analyses in order to represent the soil properties of each 20 m × 20 m plot. The analyses of soil properties were determined using methods that have been described in soil agricultural and chemical analysis (Bao, 2008). Available phosphorus (AP, mg·kg⁻¹) was measured using the molybdenum blue colorimetric method after samples had been extracted with 0.5 M Na₂CO₃. Total phosphorus (TP, g·kg⁻¹) was analyzed by acid digestion with an H₂SO₄+HClO₄ solution. Total potassium (TK, g·kg⁻¹) was digested by the HF-HClO₄-HNO₃ acid mixture and determined by the flame photometric method. Available potassium (AK, g·kg⁻¹) was shaken for 30 min with 1 M ammonium acetate solution (1:10 w/v) and then analyzed by flame photometry, and soil organic matter (SOM, g·kg⁻¹) was tested by using the dichromate oxidation method.

Trait Measurement

All woody plants (excluding vines) with a dbh ≥1 cm were sampled to measure functional traits . For each individual plant within the plots, three functional traits were measured. For each individual, we collected five well-developed and sun-exposed leaves. Leaf area (LA; cm²) was measured with a Li-Cor 3100 leaf area meter (Li-Cor, Lincoln, Neb) with leaves dried for 48 h at 65°C to determine dry mass. Specific leaf area (SLA; cm²·g⁻¹) for broadleaved species was calculated as leaf area divided by dry mass. To obtain species’ wood density (WD; g·cm⁻³), we sampled branches (1 cm ≤ diameter ≤ 2 cm) from one to ten individuals across the species in each plot. We removed the pith, phloem, and bark, measured fresh volume by water displacement and determined dry mass after drying for 72 h at 70°C (Cornwell et al., 2006). We also chose ten species of hardwood and corkwood from locations near our study site and collected branch samples and measured the stem core of ten individuals from each species (Haglof Increment Borer, Sweden). We found that the core density linearly related to the branch density for hardwood (ρcore = 1.054ρbranch, R² = 0.95, P < 0.0001) and for corkwood (ρcore = 1.235ρbranch, R² = 0.93, P < 0.0001).

Statistical Analysis

Using the data we collected, we created a dataset for trees that included the name, abundance and a measure of each of the three trait values (LA, SLA and WD) for species sampled at each of our twenty plots. Prior to statistical analyses, data for LA and SLA functional trait were log 10 transformed to reduce skew. Then, using trait-gradient analysis to decompose species trait values at two different ecological scales: alpha (within community) and beta (between community) components. The beta component refers to a measure of a species position based on average trait values for the species found in plots along a gradient, forming the x-axis of Figure 1 (the ‘trait-gradient’ of TGA). Also, it has been proposed as a measure of niche position. While, the alpha components refer to the difference between a species trait value and its beta value, (i.e. a measure of how the traits of each species differ from those of co-occurring taxa). These analyses can be expressed using the following four equations (Ackerly & Cornwell, 2007):

Figure 1.

A: Scatterplot of species trait values (t_ij) vs. abundance-weighted plot mean trait values (P_j) for log 10 SLA (cm²·g⁻¹) in Cyclobalanopsis glauca forest vegetation of Guilin karst mountainous areas. B: Distribution of trait means and regression lines for all 52 woody plant species in study.Value for two species are highlighted for illustration: Cyclobalanopsis glauca (green square) and Alchornea trewioides (yellow triangles). For each species, the large open square or triangle shows the mean position of occupied plots in which it occurs. The y-axis value for open shapes provides the mean trait value while the x-axis value provides the Beta component for the species along the trait gradient. The range of occupied plots on the x-axis is the species niche breadth (R_i), and it expresses the degree of intraspecific variation relative to the overall shift in trait values at the plot level. Alpha components reflect the deviation of each species from co-occurring taxa with regard to the trait, and α_i means the distance from the Y = X line (Y = X line, slope 1 and intercept 0 by definition). Regression line shows abundance-weighted least squares regression of species trait values relative to plot mean trait values, with slope β_i. For log SLA, t_ij varied from 2.65 to 6.96 (range = 4.31) across the entire study, while P_j ranged from 5.14 to 5.51 (range = 0.37). Each species (grey solid points) at each plot has a point-location. The vertically arranged points at a particular value of P_j represent the species that occur together in one plot.

In the equations, P_j is abundance-weighted plot mean trait values (i.e. the mean of trait values weighted by the relative abundance of each species in the plot); t_i is species mean trait values; a_ij is the abundance or weighed for species i in plot j; t_ij is species i trait value in plot j; s is the total species found within the plots; and finally, n is the total number of plots in the study.

In order to investigate the correlation between functional traits at comparable scales (the plot mean, species mean, alpha and beta levels), we calculated correlation coefficients (r) using Pearson product-moment method for each pairwise combination of the three traits. We then performed a Mantel test in the “vegan” package based on the matrix of species composition dissimilarity values and the matrix of environmental factors distances among the three field samples including 20 plots. Finally, we compared site mean trait values for species with 6 soil environmental factors. This step allowed us to arrange sites in relation to underlying environmental gradients and indicate whether trait gradients for species are aligned with particular abiotic factors (Appendix S1 of Supporting Information). All analyses were carried out in R version 3.3 (R Development Core Team, 2015) and were considered significant at the 0.05 a-level.

The Description of Parameters Characterizing

Figure 1A shows Cyclobalanopsis glauca and Alchornea trewioides as examples to illustrate the meaning of the parameters characterized in this analysis. Cyclobalanopsis glauca is a typical dominant evergreen tree species in karst mountainous areas, and occurs in almost all plots across all gradients in our study (with t_i = 4.87 and spans a range of P_j from 5.14 to 5.51) while Alchornea trewioides is a typical dominant deciduous shrub species that lives in a shaded understory in karst mountainous areas, which accounts for 14 plots of 20 plots (with t_i = 5.79 and spans a range of P_j from 5.31 to 5.51). Cyclobalanopsis glauca has low β_i and negative α_i values, reflecting the fact that it has a lower SLA than most of the other species with which it occurs (Figure 1). Species with low SLA and dense wood, like those with long-lived leaves tend to have higher survival rates because they tolerate stress from shade, wind, drought and herbivores in relatively resource-poor habitats (Chave et al., 2009). Alchornea trewioides is high β_i and positive α_i values, towards the upper of the gradient, indicating different plant life-history strategies compared with Cyclobalanopsis glauca, and reflecting an advantage in the form of a rapid acquisition of growth rate with low construction costs and rapid leaf turnover. A potential explanation for this shift towards higher SLA in Alchornea trewioides is that reduced carbon investment in structural support alters allocation patterns and allows deciduous species to dedicate a higher proportion of carbon and nutrient resources to leaf turnover. For Cyclobalanopsis glauca, R_i = P_i; for Alchornea trewioides, R_i = 5.51–5.31 = 0.20 and β_i = –2.41. Figure 1B shows the distribution of trait means and regression lines for all 52 woody plant species in study. For each species the slope of β_i was calculated by t_ij versus P_j and expresses the degree of intraspecific variation relative to the overall shift in trait values at the community level (Ackerly & Cornwell, 2007).

Correlations Between Functional Traits

Plant functional traits had different relationships within and between communities. Plot means were significantly negatively correlated between SLA and WD (Figure 2A, P < 0.01, r = –0.67). The correlation between LA and SLA was weakly positive. The species mean between SLA and WD traits showed a negative significant correlation while the correlation between LA and SLA and between LA and WD was positive (Figure 2B). For the relationship between beta components among the three traits across the plot level, they showed significant correlation (P < 0.05, r = 0.43, –0.56 and –0.85), but LA and SLA are positively correlated while LA compared to WD and SLA compared to WD were negatively correlated. The alpha components showed weaker correlations for LA and SLA when compared acrossed scales (Figure 2D); specifically, LA was positively correlated with SLA and negatively correlated with WD while SLA and WD showed significantly negative correlation (P < 0.05, r = –0.33).

Figure 2.

Scatterplots of (A) plot-mean trait values (P_j), (B) species mean trait values (C) species beta trait values (β_i) and (D) species alpha trait values (α_i) for pairwise combinations of LA, SLA and WD without log transformation. Asterisks indicate significant correlation (r) values across scales (**P < 0.01; *P < 0.05).

Partitioning of Species Trait Values

Trait variation in plant communities is thought to be constrained by two distinct and random assembly processes operating at discrete spatial scales: habitat filtering and limiting similarity between coexisting species. Species with similar functional attributes in the regional species pool are screened by the habitat filtering, leading to species with similar function being filtered into the same niche, and causing the under-dispersion of trait values within communities. On the other hand, niche differentiation prevents coexisting species from being too ecologically similar and leads to the over-dispersion of trait values, promoting species coexistence within communities (Grime, 2006). In our study, for three plant functional traits (LA, SLA and WD), the range of alpha and beta values are –1.92–2.05, –0.76–0.96, –0.53–0.37 and 3.20–3.75, 5.15–5.47, 0.99–1.08 respectively. The range of alpha components was much wider than the range of beta component for the three traits (Figure 2, Table 2), meaning a broader range of trait variation in co-existing species within shared ecological conditions. This provides a potential explanation for co-coexisting species to occupy distinct positions in multi-dimensional trait space in order to limit their similarity and reduce competition. Next, based on the calculation of LA and SLA alpha components interval (–1.92–2.05, –0.76–0.96), we found both LA and SLA were overall positive values, possibly suggesting that traits vary more within community compared with the shift across communities due to interspecific variation. However, the range of WD at the alpha components level was generally negative (–0.53–0.37), as interspecific variation will show species changing along the environment gradient. These results emphasize that a habitat filter and a limit to the similarity of coexisting species can simultaneously shape the distribution of traits and the assembly of local plant communities. In contrast, we found that a limiting similarity of coexisting species within a shared environment was greater than the habitat filtering in both LA and SLA trait value relatively. While on WD, habitat filtering contributes more to community assemblage relatively in our study.

Table 2.

Summary Statistics for Three Traits Measured in the Cyclobalanopsis glaucab Forest (LA, SLA log 10 Transformed Without WD).

Variation and Correlation of Plant Functional Traits at Comparable Scales

The partitioning of species traits into alpha and beta components provides a link between trait variation and coexistence mechanisms evaluated at comparable scales (Ackerly & Cornwell, 2007). Distinguishing between trait variation and coexistence mechanisms for a trait-based strategy is important because it gives us insight into life-history trade-offs that operate within and between environments. In the present study LA, SLA and WD showed inconsistent correlation patterns within and between communities (Figure 2). Between LA and SLA as well as between LA and WD alpha components were largely weakly correlated or not correlated across species. This appears to be consistent with earlier findings by Ackerly and Cornwell (2007), where LA varied independently from SLA in woody plant communities of coastal California, suggesting that they act as independent dimensions of variation among coexisting species. This indicates that LA and SLA strategy axes are often orthogonal in multivariate space. On the other hand, the lack of correlation in alpha components means any two species that share a site are unlikely to exhibit the same combination of trait values, and will instead exhibit a broader range of trait variation, reflecting the differentiation of niche space along dimensions of varying habitat conditions (Rachael & Michelle, 2012). WD was largely unrelated to the other traits across species except for exhibiting a negative correlation with SLA. However, significant correlations were found at beta components for each other, indicating correlations between these two trait pairs were predominantly driven by their tendency to vary in parallel along wider scale abiotic gradients, though alpha components contributed more variation than beta components (Kooyman et al., 2010). In community ecology, the convergence of trait function in relation to edaphic and climatic conditions is viewed as evidence for habitat filtering (Cornwell et al., 2006; Kooyman et al., 2010). For SLA and WD, there were significant negative correlations (across species, alpha, beta level and plot means). However, evidence for these two traits have been different across species. Santiago et al. (2004) found wood density and photosynthetic capacity (A mass) to be negatively correlated. While two studies found that specific leaf area and wood density were negatively correlated (Bucci et al., 2004; Ishida et al., 2008), another found no correlation (Wright et al., 2007) . In our study, the negative correlations between them are the same with those reported Bucci et al. (2004) and Ishida et al. (2008) across species. Generally, no correlations or weak correlations were found among traits at the alpha level. However, a negative relationship was presented between WD and SLA not only at the alpha level but also at the rest of the scales. This may be related to the specific habitat of our field sampling. Because the samples were carried out in the late of succession for the Cyclobalanopsis glauca forest, evergreen species occupied a large proportion compared to deciduous species. Furthermore, we found species with small SLA and higher WD in a large proportion of most plots and species with this combination of traits were more abundant in some plot relative to others. This results in slow-growing species with high resource-use efficiency which gradually became the dominant species in a later successional stage of the Cyclobalanopsis glauca forest. An explanation for this may be that species were negatively correlated at comparable scales.