Study site

Xishuangbanna Dai Autonomous Prefecture (21°08′ – 22°35′N, 99°56′ – 101°50′E) borders Myanmar in the southwest and Laos in the southeast (Figure 1). The region has a mountainous topography, with mountain ridges running in a north–south direction and a southward decrease in elevation. It has a typical monsoon climate with distinct wet and dry seasons. Mean annual air temperature and rainfall are 21.8°C and 1493 mm, respectively. Approximately 80% of the annual rainfall occurs in the wet season, which lasts from May to October. The frequent occurrence of heavy fog in the dry season in the lower hills and valleys may assist the expansion of the northern limit of tropical rain forest to this region (Cao et al 2006).

FIGURE 1 

Location of 20 plots in 4 elevational bands (circled) in Xishuangbanna National Nature Reserve, southwest China. (Map by authors)

Data collection

In August 2012, we established a transect consisting of 4 elevational bands (800, 1000, 1200, and 1400 m above sea level—Figure 1 and Table S1, Supplemental material,  http://dx.doi.org/10.1659/MRDJOURNAL-D-15-00109.S1). in the Xishuangbanna National Nature Reserve. Elevations below 800 m above sea level were not included in this study, as the forests were cleared or heavily disturbed. Within each elevational band, we set up 5 replicate plots of 20 m × 20 m that were spaced approximately 200 m from each other. Locations impacted by anthropogenic and natural disturbances, such as large canopy gaps, were avoided. Within each plot, all trees with a diameter at breast height ≥5 cm were measured and tagged. Most trees were identified on-site by local botanists; those that could not be identified were collected and brought back to Xishuangbanna Tropical Botanical Garden herbarium for further determination. In December 2013, five 1 m × 1 m subplots were established at the 4 corners and the center of each 20 m × 20 m plot to survey seedlings. Within each subplot, all stems of tree species <1 cm diameter at breast height (hereafter referred to as seedlings) were tagged and identified by species.

We recorded hourly temperatures from May 2013 to April, 2014 using a thermo-logger (DS1923 Hygrochron iButton, Maxim, CA, USA) set at the height of 1.3 m in the vicinity of each plot. Due to technical problems, no temperature measurements at 1200 m were made in July and August, and thermo-loggers at three 1000 m plots were lost during the dry season. We measured soil moisture at 5 cm below the ground at 5 randomly selected points within each subplot (a total of 25 measurements per plot) using a conductivity probe (Theta probe MPM-160B, ICT International Pty Ltd, Armidale, New South Wales, Australia) at the end of the wet (December 2013) and dry (April 2014) seasons. We used a geologic compass (HRB DQY-1, Harbin Optical Instrument Factory, Heilongjiang Province, China) to measure the slope and aspect of each plot.

Data analysis

The seedling data collected from the five 1 m × 1 m subplots in each plot were pooled before analysis. The abundances of the adult trees and the seedlings were log-transformed before calculating 3 species diversity indices: the Shannon-Wiener index, Simpson index, and Pielou evenness index (Magurran 1988).

To test whether diversity and species richness differed in the different elevational bands, we used one-way analysis of variance (ANOVA) with post hoc Tukey's tests. Differences in assemblage composition were investigated using nonmetric multidimensional scaling (NMDS) ordinations with 10 random restarts (Minchin 1987), based on 4 types of similarity matrices: the Jaccard, Bray-Curtis (using log-transformed abundances), Chao-Jaccard (Chao et al 2005), and Raup–Crick (Chase et al 2011) indices. All species that occurred in 1 site only were removed before calculating similarity values to reduce overestimation of beta-diversity. Permutational multivariate analysis of variance (PERMANOVA) was used to test if the species compositions of adult trees and seedlings were significantly different among the 4 elevational bands (Anderson 2001; McArdle and Anderson 2001).

We adopted an information-theoretic approach (Burnham and Anderson 2002) to test whether assemblage composition of adult and seeding species was related to temperature, soil moisture, or other environmental factors, including elevation and geographic factors. We generated 12 environmental variables, which were grouped into 5 classes with similar attributes (temperature, soil moisture, elevation, topography, and spatial arrangement). For temperature data, we first calculated the average air temperature during the day (7:00 to 18:00 h) and at night (19:00 to 6:00 h) of each day in January–April (the dry season). These values were then used to calculate mean, maximum, and minimum air temperatures and their coefficient of variation. Wet season temperature data were not used due to the lack of data at 1200 m (explained earlier) and negligible differences between elevational bands in this season (Figure 2). Soil moisture measurements within each plot were averaged to generate wet-season (December) and dry-season (April) soil moisture and its seasonal variation. Topography consisted of slope and aspect measured within each plot. Spatial arrangement of the plots was quantified using the primary and secondary vectors of the principal coordinates of neighbor matrices (PCNM1 and PCNM2), which showed positive spatial correlation based on Moran's I (Dray et al 2006; Legendre and Legendre 2012).

FIGURE 2 

Mean monthly air temperature in the 4 elevational bands, May 2013 to April 2014. Data for the 1200 m band in July and August are missing (bars represent standard error).

We employed a model averaging method (Burnham and Anderson 2002) to investigate the relative importance of individual predictor variables in explaining variation in the composition of adult tree and seedling species. Using multivariate generalized linear models developed by Wang et al (2012), we selected plausible predictor variables by testing whether the sum of the Akaike weights of each predictor variable was significantly greater than the summed Akaike weights obtained from a null distribution of the plots. The null model was generated by permuting plots (not individuals), as this preserved the tree species composition, and thus species interactions, of each plot. We compared the observed Akaike weights with those derived from 999 permutations. Due to the loss of temperature thermo-loggers from 3 plots at 1000 m during the dry season, we included only 17 plots (instead of 20) for the analysis using all 12 predictors. We then re-ran the same analysis using all 20 plots but excluding temperature predictors. Abundances were square-root transformed before analysis (Nakamura et al 2015).

To investigate differences (and similarities) in the elevational distributions of adult tree and seedling species, we used the indicator value protocol developed by Dufrêne and Legendre (1997), using the labdsv R package version 1.6-1. We calculated the indicator value of a given species for single elevational bands (eg 800 m, 1200 m) and all possible sequential elevational bands (eg 800–1000 m, 1000–1400 m) omitting the entire elevational range (ie 800–1400 m), giving a total of 9 indicator values per species. A random reallocation procedure with 4999 permutations was performed to test for the significance of indicator values. We selected the species with significant indicator values greater than 60% (a subjective benchmark value). We also calculated the importance value, which additively summarizes the relative abundance and the relative frequency of the selected adult and seedling tree species in each of the elevational bands (Curtis and McIntosh 1951).

Diversity and species composition across elevations

Individual-based rarefaction curves showed greater observed and estimated (by extrapolation) species richness at lower (800 and 1000 m) than higher (1200 and 1400 m) elevational bands for both adult trees and seedlings (Figure 3). Within each of the lower and higher elevational bands, the differences in species richness were indistinguishable, with overlapping 95% confidence intervals. This was also statistically confirmed, with the results of ANOVA showing both adult trees and seedlings having higher species richness within lower elevational bands (F = 67.56 for adult trees, F = 39.77 for seedlings; see also Figure 4).

FIGURE 3 

Nonmetric multidimensional scaling ordinations in the 4 elevational bands based on different similarity indices: (A) Bray-Curtis; (B) Chao; (C) Jaccard; (D) Raup-Crick.

FIGURE 4 

Mean species richness of seedlings and adult trees in each elevational band. Lowercase letters indicate differences between elevational zones. Bars indicate standard errors.

Regardless of the similarity indices used, NMDS ordinations showed clear elevational stratification in both adult tree and seedling species composition (Figure 3). PERMANOVA confirmed significant elevational differences (P < 0.001) for both adult tree and seedling species composition. All of the ordinations showed a notable gap between the low elevations (800 and 1000 m) and the high elevations (1200 and 1400 m), and the sample dispersion was generally greater at the low elevations.

Temperature and soil moisture

Mean monthly temperature showed little difference in May–August among the elevational bands (Figure 2). Although the temperature was highly variable in September–December, it was generally warmer in the lower elevational bands. In contrast, temperatures in January–April were cooler at 800 m than at other elevational bands.

We found significant differences in soil moisture among the elevational bands in both wet and dry seasons (F = 7.109 and F = 13.88 in wet and dry seasons, respectively). Post hoc tests showed that the wet season soil moisture was significantly greater at lower (800 and 1000 m) elevational bands (Figure 5). This pattern is consistent with those of adult and seedling species richness (Figure 4) and assemblage composition (Figure 3). In the dry season, only the soil moisture at 800 m was significantly higher, and soil moisture at 1000 m was not different from that at 1200 and 1400 m.

FIGURE 5 

Soil moisture at 4 elevations at the end of the wet and dry seasons. Lowercase letters indicate differences in soil moisture. Bars indicate standard errors.

Relationships with environmental factors

For both partial (17 plots with 12 predictors) and full (20 plots with 8 predictors) datasets, the results of the information-theoretic approach suggested a number of plausible predictors that may explain the assemblage composition of adult and seedling species. Due to the spatially aggregated nature of the elevational transect, the spatial predictor (PCNM1) was plausible for both adult trees and seedlings (Tables 1 and 2). Elevation was also a plausible predictor for both adult trees and seedlings, with the exception of adult trees when using the full dataset (20 plots). No environmental predictors were selected for seedling assemblage composition, whereas mean temperature and dry season soil moisture were significantly associated with adult trees when using partial and full datasets, respectively.

TABLE 1 

Results of the information-theoretic analysis of assemblage composition of adult trees using 17 plots with all predictor variables.

a)

Summed Akaike weights use the model averaging approach.

b)

Significant P values are shown in bold.

c)

PCNM1 and PCNM2 are the primary and secondary vectors of the principal coordinates of neighbor matrices.

TABLE 2 

Results of the information-theoretic analysis of assemblage composition of adult trees using all 20 plots without temperature predictors.

a)

Summed Akaike weights use the model averaging approach

b)

Significant P values are shown in bold.

c)

PCNM1 and PCNM2 are the primary and secondary vectors of the principal coordinates of neighbor matrices.

Elevational distribution of adult trees and seedlings

We found 20 and 16 significant indicator species for adult trees and seedlings, respectively (Table 3). Of these, 19 were indicative of only 1 elevation (10 and 12 for adult trees and seedlings, respectively), 15 were indicative of 2 elevations (10 and 5 for adult trees and seedlings, respectively), and none was indicative of 3 or 4 elevations. All indicator species were restricted to either low (800 and 1000 m) or high (1200 and 1400 m) elevations. The only exception was Aporosa yunnanensis, whose adult trees and seedlings indicated elevations of 1000 and 1200 m. When both the adult trees and the seedlings of the same species were found to be indicators, their indicated elevations generally matched, with the exception of C. echinocarpa, whose seedlings showed a narrower elevational range than the adult trees.

TABLE 3 

Elevation-restricted species with significant indicator values of 60% or higher.^a) (Table continued on next page.)

a)

* Indicates that the adult tree or seedling was not a significant indicator of any elevational band; – indicates that the adult tree or seedling was not present at any elevational band.

Diversity and species composition across elevations

We found a marked decrease in species richness from 1000 to 1200 m for both the adult tree and the seedling assemblages (Figure 4 and Table 4). Such a step-wise change in diversity patterns was also reported from other tropical mountains in Malesia (Van Steenis 1984), Borneo (Aiba and Kitayama 1999), Bolivia (Kessler 2000), Tanzania (Hemp 2006), and the Caribbean (Martin et al 2011).

TABLE 4 

Species diversity indices for adult trees and seedlings at different elevations. Lowercase letters indicate significant differences between elevation zones.

Furthermore, we found a clear division in assemblage composition between low (800 and 1000 m) and high (1200 and 1400 m) elevations (Figure 3). This division was also evident in the indicator species, where only 1 species (A. yunnanensis) was indicative of both low and high elevational bands. Our results were consistent with previous research showing that large members of tropical tree species were restricted to a single forest type (Pitman et al 1999). The dominant tree species at low and high elevations belong to the genera Parashorea and Castanopsis, respectively (Table 3). These findings were consistent with previous studies based on adult trees (Wu et al 1987; Ohsawa 1993; Zhu et al 2006). The relative narrowness of the elevational distributions of dominant tree species is caused by opposing environmental constraints (Williams et al 2010; Arellano and Macía 2014). The vegetation changes observed in our study were likely influenced by strong seasonal dryness (and perhaps temperature—discussed later), with tropical rain forest occurring in wet valleys and tropical montane evergreen broad-leaved forest occurring in dry montane habitats (Zhu et al 2005).

Species composition was more similar for both adult trees and seedlings at high elevations (Figure 3). This may have resulted from drier environmental conditions at high elevations acting as an environmental filter and reducing species richness (Zhu et al 2005; Cao et al 2006). Aggregation of high-elevation plots was observed in NMDS ordinations of all 4 similarity indices, including the Raup Crick index (a probabilistic measure unaffected by the number of species— Chase et al 2011), suggesting that the high similarity of these plots is not a simple artifact caused by depauperate flora at these elevations.

Associations with environmental factors

Elevation is a suitable proxy for varying environmental factors, as the assemblage composition of organisms and environmental characteristics generally change with elevation in a progressive manner (with some exceptions—eg Salas-Morales and Meave 2012). Both adult tree and seedling compositions in our study, however, showed a clear division between high and low elevations, suggesting that some important environmental factors might change significantly across these elevations.

For adult trees, we found that mean temperature and dry-season soil moisture were plausible predictors, significantly associated with assemblage composition. Temperature, generally speaking, decreases linearly with elevation (Barry 1992). Our results, however, showed much lower mean temperature at 800 m (17.5 ± 0.34SE) than at higher elevations (20.0 ± 0.02, 19.4 ± 0.18, and 18.4 ± 0.07 at 1000, 1200, and 1400 m, respectively) in the dry season (Figure 2). Several of the previous surveys in this area reported similar patterns in temperature (Guo and Zhong 1986; Zhong and Guo 1997). As the 800 m plots were all located within a mountain valley, our results may be attributable to cool air draining down into the valley and lifting the warm air to higher elevations (Miller et al 1983; Goulden et al 2006).

However, our results have caveats because the analysis included partial data (n = 17), which excluded 3 of the 1000 m plots due to loss of the thermo-loggers. Using the full 20 plots, we found that soil moisture in the dry season was a plausible environmental factor (Table 2). Soil moisture was significantly higher at 800 m than at other elevational bands at the end of the dry season, perhaps due to the presence of heavy fog in the lower valley (Cao et al 2006). Although similar results were found during the wet season, where soil moistures at both 800 and 1000 m were significantly higher than at 1200 and 1400 m, wet season soil moisture was not selected as plausible. This implies that limited soil moisture availability during the dry season may shape the elevational distribution of adult trees in this area. Water shortage has been associated with the distribution of tree species in many parts of the world. For example, in tropical montane forest, gullies and high elevations with light water stress are dominated by humid-condition species, while low elevations are dominated by dry-condition species (Ledo et al 2009). In subtropical forest, soil moisture drives change in elevational vegetation types (You et al 2013). This could even be aggravated in mountains dominated by monsoon climate. In the temperate forest, drought resistance ability significantly influenced the elevational positions of 5 pine species (Barton and Teeri 1993). In Panama, soil water availability was considered to be the main factor controlling the distribution of tropical tree species (Bunker and Carson 2005; Engelbrecht et al 2007), perhaps because the survival and growth of seedlings is constrained by the severity of drought in the dry season (Engelbrecht et al 2005; Slot and Poorter 2007; Comita and Engelbrecht 2009). Our study suggested that moisture availability and temperature regime, which are caused by the change of elevation, may determine the elevational distribution pattern of adult tree species.

We found that seedling composition, in contrast to that of adult trees, was only associated with elevation and spatial arrangement. Soil moisture and temperature were not selected as plausible predictors of seedling composition (Tables 1 and 2). The lack of plausible environmental predictors for seedling assemblage composition may suggest that environmental filtering only operates while trees are maturing (Grubb 1977; Poorter et al 2008). Our result suggests that the seedlings of high-elevation tree species established at low elevations may play an important role in shifting species ranges to stay in a suitable environmental envelope.

Elevational distributions of adult trees versus seedlings

We observed very similar elevational distributions for adult trees and seedlings of most indicator species (Table 3). Zhu et al (2014) also reported that most tree species in North America are not yet migrating in response to climate change. In our study, we found several tree species that dominate the evergreen broad-leaved forest at 1200 and 1400 m, such as C. echinocarpa and C. calathiformis, occurring at lower elevations where the tropical rain forest is distributed. However, we did not observe seedlings of tree species dominant at 800 m and 1000 m distributed in higher elevations. Under potential climate change regimes, these changes in the elevational distribution of seedlings may expand the range of tree species in the future and ultimately alter forest structure and processes. This suggests that the species currently dominant at high elevations might expand their range downward if future climate change causes habitats at lower elevations to become more suitable for their establishment (eg dryer). However, under the same climate change scenarios, species currently dominant at low elevations are unlikely to expand upslope because their seedlings are less adapted for the dryer habitats found at high elevations.

In a climate warming scenario, shifts in distribution range of seedlings may lead to shifts in the distribution range of tree species, which ultimately influences the structure and dynamics of the forest (Lenoir et al 2009; Rehm and Feeley 2013). Previous studies indicated that some species in temperate areas shifted their distribution to higher elevations because of climate warming (Wilson et al 2005; Lenoir et al 2008). In some tropical forests, climate warming is considered to be the main factor driving tree species to migrate upslope (Feeley et al 2013; Savage and Vellend 2015). However, such a situation may not take place in our study region, as the relatively dryer conditions at higher elevations may limit the uphill shift of tree species in tropical rain forests. Our results suggest that, as this region is predicted to become drier and hotter with climate change (He and Zhang 2005), the tropical rain forests at lower elevations may contract their distribution ranges as the habitats at lower elevations become drier, and species at higher elevations may shift their distributions downward.

Habitat modification can also be an alternative mechanism causing downslope range shifts (Lenoir et al 2010). A study in California suggested that species shift downhill in response to regional changes in climatic water balance (Crimmins et al 2011). Furthermore, our research region may face a significant loss of species diversity if tree species cannot shift their elevational distributions as quickly as climate change occurs (Corlett and Westcott 2013). Some tree species of the Fagaceae family may expand their local distribution. High-elevation species such as C. echinocarpa and C. calathiformis could likely shift to low-elevation areas, but the representative tree species of low-elevation tropical rain forests, such as P. chinensis and Baccaurea ramiflora, may narrow their distribution range. Under climate change regimes, this downhill migration of high-elevation species might also occur in other tropical and subtropical montane forests dominated by monsoon climate.