Direct and indirect effects of environmental factors, spatial constraints, and functional traits on shaping the plant diversity of montane forests

Abstract Understanding the relative importance of the factors driving the patterns of biodiversity is a key research topic in community ecology and biogeography. However, the main drivers of plant species diversity in montane forests are still not clear. In addition, most existing studies make no distinction between direct and indirect effects of environmental factors and spatial constraints on plant biodiversity. Using data from 107 montane forest plots in Sichuan Giant Panda habitat, China, we quantified the direct and indirect effects of abiotic environmental factors, spatial constraints, and plant functional traits on plant community diversity. Our results showed significant correlations between abiotic environmental factors and trees (r = .10, p value = .001), shrubs (r = .19, p value = .001), or overall plant diversity (r = .18, p value = .001) in montane forests. Spatial constraints also showed significant correlations with trees and shrubs. However, no significant correlations were found between functional traits and plant community diversity. Moreover, the diversity (richness and abundance) of shrubs, trees, and plant communities was directly affected by precipitation, latitude, and altitude. Mean annual temperature (MAT) had no direct effect on the richness of tree and plant communities. Further, MAT and precipitation indirectly affected plant communities via the tree canopy. The results revealed a stronger direct effect on montane plant diversity than indirect effect, suggesting that single‐species models may be adequate for forecasting the impacts of climate factors in these communities. The shifting of tree canopy coverage might be a potential indicator for trends of plant diversity under climate change.


| INTRODUC TI ON
Understanding the relative importance of the factors driving patterns of biodiversity is an important topic in ecology and biogeography (Gaston, 2009). However, we still do not have a thorough understanding of the factors limiting the patterns of plant community diversity (Heino & Tolonen, 2017;Victorero, Robert, Robinson, Taylor, & Huvenne, 2018), especially in mountain regions over the world (Gaston, 2009).
Plant communities are complex and composed of organisms with very different life history traits, thermal tolerances, and dispersal ability (Classen et al., 2015). Therefore, the diversity of plant communities is not only controlled by external factors (i.e., geography, land cover, and environmental conditions) (Victorero et al., 2018;Xiong et al., 2019), but also by internal factors (biological characteristics, such as life history and functional traits) (Soininen, Lennon, & Hillebrand, 2007). Most studies have generally agreed that variation of environmental factors (including climate change) dominates deterministic processes of plant community composition changes (Heino, Mykrä, Kotanen, & Muotka, 2010;Victorero et al., 2018). However, some studies have pointed out that there are significant correlations between species characteristics/ traits (i.e., biomass, canopy height, and leaf area) and plant community diversity (Jenni, Janne, & Helmut, 2010;Soininen et al., 2007).
Moreover, the effect of environmental changes on species diversity is mediated by plant functional traits (Heino & Tolonen, 2017;Mcgill, Enquist, Weiher, & Westoby, 2006;Verberk, Noordwijk, & Hildrew, 2013). Thus, conducting a study on functional traits may contribute to understanding how environmental conditions filter species from regional species pools and how species compete for resources.
Environmental factors, biological factors, and spatial constraints are interrelated or change together under natural conditions (Mcgill et al., 2006;Verberk et al., 2013). These multiple drivers can interact in ways not predictable by single factor effects to directly influence vital rates of plant diversity through nonadditive effects on demography, physiology, and morphology (Farrer, Ashton, Knape, & Suding, 2014). Indirect effects can be driven either by changes in the abundance of other species or by changes in the direction and/or strength of per capita interaction effects via functional traits (Gilman, Urban, Tewksbury, Gilchrist, & Holt, 2010;Tylianakis, Didham, Bascompte, & Wardle, 2008). For example, climate change may lead to an increase in abundance of one plant species by reducing the abundance of another (reducing competition) (Li et al., 2020). In addition, these biotic interactions may covary with environmental gradients, further confounding our understanding of the true strength of the abiotic and biotic drivers (Chu et al., 2019). Given the potential importance of indirect effects, ignoring biotic interactions could severely affect the accuracy of forecasts of species abundances and distributions under a changing environment (Angert, LaDeau, & Ostfeld, 2013;Chu et al., 2019), consequently limiting the effectiveness of conservation and management actions. However, few studies have addressed the importance of direct effects (single factor or interactive) and indirect effects of driving factors on the species diversity and distribution of plant communities. Indirect effects of environmental variables on plant community can be strong in High Arctic or alpine ecosystems (Dormann, Wal, & Woodin, 2004;Klanderud & Totland, 2005), whereas direct effects can predominate in other ecosystems (Levine, McEachern, & Cowan, 2010). Alternatively, mechanisms might be largely site-specific, varying in both direct and indirect drivers owing to unique climatic conditions and management history.
Vegetation composition and patterns in montane forests may be influenced by direct and/or indirect drivers of different environmental factors, geography, and biological characteristics (Chu et al., 2019;Li et al., 2020). Montane forests are characterized by high topographic variability in a small area; this variability includes topographical factors such as elevation, slope inclination, and ground surface texture (Andrus, Harvey, Rodman, Hart, & Veblen, 2018). In addition, the vegetation in montane forests is highly sensitive to climate change with multiple functional traits Xiong et al., 2016).
Because of these abiotic and biotic differences, plant diversity might vary in terms of their responses to multiple drivers, especially when studied as a network of indirect and direct drivers of diversity. With field data from a multisite survey, we identify direct and indirect pathways linking multiple drivers to plant diversity in montane forests. We hypothesize that (a) spatial and climate factors, such as mean annual temperature (MAT), precipitation, and spatial distance, are the main drivers of plant species diversity; (b) functional traits play a small but significant role in plant diversity; and (c) overall responses to spatial and climate drivers are driven primarily by direct effects. The aims of this research were to (a) identify the main factors influencing the interrelationships between environmental, spatial, and functional traits, and explore the causes that lead to a change in plant diversity and (b) disentangle the direct and indirect effects of environmental, spatial, and functional traits on species diversity within a plant community in montane forests. By characterizing the complexity of diversity responses to multiple drivers, we aim to contribute to a predictive understanding of how direct and indirect effects drive the variability of plant diversity in montane forests.

| Study area
This study was conducted in the Sichuan Giant Panda habitat, which is located in an alpine valley in the transition region from the Qinghai-Tibetan Plateau to the Sichuan basin (Li et al., 2019). The area is part of the subtropical evergreen broadleaf forest region and warm temperate deciduous broadleaf forest region (Sichuan Vegetation Cooperation Group, 1980). The Sichuan Giant Panda habitat was established as a UNESCO World Heritage Site in 2006. It is a refuge to diverse wildlife and plant species, as well as home to more than 30% of the wild giant panda population (State Forestry Administration, 2006).
In fact, the region is within one of the world's top 34 biodiversity hotspots (Bellard et al., 2014) and one of the Global 200 Ecoregions defined by the World Wildlife Fund (WWF) . Within each plot, trees in a 20 m × 30 m subplot and shrubs from three 5 m × 5 m subplots were studied. Data from the three subplots within each plot were then pooled. The plant species, number of individuals (abundance), and coverage of each layer (e.g., tree, shrub) were recorded (Table S1).

| Variables of environmental, spatial constraints, and plant functional traits
A total of five environment factors were recorded for each sample (Li et al., 2019): altitude, slope, aspect, mean annual temperature (MAT), and precipitation. Meteorological data were mainly obtained from in situ monitoring in each protected area.
Variables of spatial constraints mainly include longitude, latitude, and spatial distance (Li et al., 2019). The spatial variables that constrain the ordinate model were provided based on Moran's eigenvector (MEM) (Dray, Legendre, & Peres-Neto, 2006). In order to avoid spatial autocorrelation, we used Moran's I index to extract 90 spatial variables with positive eigenvalues and negative correlations.
The MEM spatial variable was obtained by the Principal Coordinates of Neighbor Matrices (PCNM) function in the PCNM package. The spatial distance between the sample sites was calculated by the longitude and latitude in the geosphere package in R (Legendre, Borcard, & Peres-Neto, 2012).
Seven plant functional traits, namely tree canopy coverage, diameter at breast height (DBH), canopy height, specific leaf area, leaf area (LA), leaf thickness (LT), and leaf dry matter content (LDMC), were measured at each study site (Li et al., 2019). These traits reflect the substance exchange balance between plant resource acquisition and protection under environmental change (Bernard-Verdier et al., 2012).
We screened mature plant individuals without pests and diseases. For each species, we randomly selected five individuals and repeated sampling five times for each selected individual (Pérez-Harguindeguy et al., 2013). The functional trait data were obtained from those samples in the plot. The functional traits of the main dominant tree and shrub species were presented as a weighted average, which was used to investigate the functional traits of the plant community as a whole. We did not establish the relationship between climatic factors and topographic factors (elevation, slope, and slope direction), because climatic factors had limiting affect on topography in decades in this area . In addition, we hypothesized that climatic variables would significantly affect plant functional traits and preserve the inclusion of functional traits (p < .05) in the optimal model.

If climatic factors did not significantly affect plant functional traits
or if their addition led to a decrease in the best model interpretation, we deleted the correlation between climatic factors and plant functional traits. We used the "lavaan" package in R to model the structural equation (Rosseel, 2012).
Model evaluation was determined by the chi-square (χ 2 ) test (p > .05 for a satisfactory fit) and the standardized root mean square residual (SRMR < 0.05 for a satisfactory fit). The Akaike information criterion (AIC) was used to select the best model with a satisfactory fit. When a model met the criteria of the chi-square test and SRMR but contained nonsignificant paths, we repeated the modeling fit and evaluation by removing these paths. Therefore, the final selected model may not have a minimum AIC value (Li et al., 2019). The decision to remove a path was primarily based on the p-value for the path and the performance of the overall fit of the model. The total effect that one variable had on another equaled the sum of its direct and indirect effects through directed (causal) paths. The SE values and p-values for standardized path coefficients were obtained through the function standardized solution in the "lavaan" package of R.
To further understand the response of plant alpha diversity to selected functional traits, climate change, and altitude from SEMs, we analyzed the changes in richness and abundance of shrubs, trees, and the overall plant community through a general linear model (GLM) with a Poisson family distribution using the selected functional traits, climate change, and altitude factor as the dependent variable and species richness/abundance as the response variable.

| Environmental and spatial constraints affecting plant community composition
The  impact factor (environmental factors, spatial constraints, and functional trait) matrices (r = .13, p value = .001) (Li et al., 2019). There was a significant correlation between tree species diversity and environmental factors (r = .10, p value = .001) and a significant correlation between tree species diversity and spatial constraints (r = .32, p value = .001). Moreover, the species diversity of shrubs was significantly correlated with the total impact factor matrix (r = .20, p value = .001) (Li et al., 2019), spatial matrices (r = .22, p value = .001), and environmental matrices (r = .19, p value = .001). In addition, there was a significant correlation between total plant species diversity and total impact factor matrices (r = .23, p value = .001). There was a significant correlation between plant community species matrix, spatial matrix (r = .36, p value = .001), and environmental matrices (r = .18, p value = .001) (Li et al., 2019). There was no significant correlation of trees, shrubs, or overall plant diversity with the plant functional trait matrix.

| Direct and indirect effects of spatial, environmental, and functional traits on plant diversity
In total, there were 233 shrub species and 174 tree species in the field sites. Because the spatial distance is autocorrelated with the latitude and longitude, we excluded the spatial distance MAT and annual precipitation did not significantly affect tree species richness (Figure 3). Latitude (Z value = 2.275, p = .023) and the coverage of the tree canopy (Z value = 3.54, p < .001) significantly affected tree richness. MAT (Z value = 7.76, p < .001), the coverage of the tree canopy (Z value = 14.33, p < .001), and latitude (Z value = 9.23, p < .001) significantly affected tree abundance, while precipitation did not affect tree abundance (Figure 3).

| Spatial and environmental variables primarily drive the spatial change of plant diversity
Studies support that the pattern of biodiversity of an ecosystem is influenced by a variety of local to regional factors Mykrä et al., 2010). The importance of these factors for biodiversity patterns may depend on the spatial dimension of the study area and the characteristics of the species diversity . Climate is a key driver of the composition of subalpine plant communities (Pauli, Gottfried, Reiter, Klettner, & Grabherr, 2007;Xiong et al., 2016). Our results indicated that MAT and precipitation are important influencing factors that can describe the spatial shifting of plant species diversity; in addition, climate variation can be used as a predictor for the pattern of diversity of montane plant species. Relevant studies have proposed that climate has a strong filtering effect on plant communities (Soininen et al., 2007). In our study, although climate factors had low explanatory power for plant communities (Figure 1), it should be noted that other environmental variables, such as soil moisture, pH, and total nitrogen, have often been found to be critical in determining plant community composition in subalpine montane forests (Hettenbererova, Hajek, Zelený, Jirouskova, & Mikulaskova, 2013). More explanatory variables may be needed to account for the changes in species diversity.
We found that environmental variables and spatial constraints largely explained the diversity of trees, shrubs, and plant commu-  (Leibold et al., 2010). According to previous studies, spatial factors may be related to diffusion restrictions and could play an important role in determining plant species diversity on a broad spatial scale . Therefore, considering the spatial area covered in our study, the decentralized restriction on the regional scale may also drive the change of plant diversity to some extent. In addition, spatial distance and altitude significantly affect plant species diversity, which also indicates that spatial factors limit the ability of plants, especially tree species, to spread over a great Some studies have suggested that biotic interactions may be just as influential in shaping plant community diversity and composition (Warren & Bradford, 2011). However, our findings showed that functional traits may not be as strong drivers of diversity in montane forests as in grasslands or mesic forests (Warren & Bradford, 2011;White, Bork, & Cahill, 2014). The use of plant functional traits may allow for more informative comparisons with regard to gauging ecosystem integrity (Warren & Bradford, 2011). and species abundance are strongly correlated, which is consistent with the theory of plant resource acquisition (Heino & Tolonen, 2017

| Direct and indirect effects of driving factors on plant community diversity
In our study, SEMs were employed to distinguish direct and indirect effects of environmental factors and biotic interactions on the dynamics of plant diversity. The diversity (richness and abundance) of shrubs, trees, and plant communities was directly affected by precipitation, latitude, and altitude. However, MAT has no direct effect on the richness of trees and plant communities. As previously reported, the effects of temperature on species richness are direct rather than indirect in grasslands (White et al., 2014). Moreover, Hoeppner and Dukes (2012) reported negative responses of richness to warming and provided evidence indicating resistance of grassland diversity to the direct effect of warming. This may be due to the delayed response of perennial trees to temperature changes (Xiong et al., 2016). Moreover, it has been found that trees and plant communities in montane forests are not sensitive to temperature changes over a 10,000-year time scale , indicating that warming is not the most citing precipitation and latitude as important for the dynamics of plant diversity Mykrä et al., 2010). Specifically, we found precipitation and latitude consistently important in all three models (Figure 2).
Our results indicate that environmental factors have direct effects related to increasing plant community species richness and abundance. Precipitation has been reported to promote the richness of trees and shrubs, and increasing MAT elevated the richness of shrubs in subalpine mountains as well as grasslands (Chu et al., 2019;Lin, Xia, & Wan, 2010;Xiong et al., 2019). The results from the SEMs showed that MAT may decrease the tree canopy and indirectly inhibit tree diversity (Figure 2) (Chu et al., 2019). For the differences in the direct and indirect responses of trees and shrubs to environmental factors, ecological niche differences may influence the magnitude of indirect effects. Another factor contributing to variability in the size of raw indirect effects is asymmetry in interspecific interactions (Kleinhesselink & Adler, 2015). The results also illustrate the diversity of species' responses to environmental variation.
However, the relative importance of indirect effects to direct effects could change across the range of a species. Some studies have reported that indirect effects of climate change can amplify, outweigh, or even reverse direct effects (Suttle, Thomsen, & Power, 2007;Tylianakis et al., 2008).
Our findings also indicate some promising future directions.
First, the potential importance of indirect effects (the mediating effect of functional traits and biotic interactions) was ignored.
Solely focusing on the direct effects of biotic interactions could lead to a severe underestimation of the effect on species abundances and distributions under a changing climate (Angert et al., 2013), consequently limiting the effectiveness of conservation and management activity. Second, the inclusion of additional functional traits, such as foliar profile, representing the vertical dimensions of forest structure, remains a promising area for additional studies. Further, the considerable unexplained variance found in this study suggests that other unmeasured factors (e.g., the abundance of herbivores and pathogens, soil properties) may play a greater role in determining species richness in these forests (Xiong et al., 2016.

| CON CLUS IONS
By using a framework capable of identifying both direct and indirect responses, we determined the primary drivers of plant richness and abundance in montane forests: climatic factors and spatial constraint (latitude). Our results demonstrate that interactions among environmental factors, spatial constraints, and functional traits both directly and indirectly influence plant species richness in montane forests. The correlations of trees, shrubs, or plant composition with the environmental and spatial constraints were significant, while there was no significance for the plant functional traits. Spatial constraint variables were the main driving force shifting plant species diversity.
Moreover, the diversity of shrubs, trees, and plant communities was directly affected by precipitation, latitude, altitude, and the tree canopy. However, MAT had no direct effect on the richness of trees and plant communities. MAT and precipitation via the tree canopy indirectly affected tree richness and abundance. Our results also found that the direct effect was significantly stronger than the indirect effect. Precipitation generally had a positive relationship with plant richness. These findings show that a number of mechanisms act in concert to shape the environmental gradient related to plant diversity, with no single mechanism being sufficient on its own. Our results also illustrated the complexity of ecosystem responses that could unfold following seemingly simple modifications of single factors. However, the factors underlying variability between systems may contribute to predicting how systems will respond, which can be identified by understanding the key drivers of system responses. These findings show that the impact of climate change on plant diversity might be indirectly predicted by the effects of climate change on tree canopy coverage. We appreciate the insightful observations of Associate Editor and two anonymous reviewers of previous versions of this article.