Results
Trait- environment interactions: As expected, the four traits differed in their interaction with environmental variables to predict species distributions across the landscape (Krishnadas et al., 2018b), and trait-environment interactions differed between fragments and contiguous forest (Table 1). Interestingly, models with elevation alone explained similar extents of variation in trait-mediated abundances as models with multiple environmental predictors (Table S1). In contiguous forest, higher elevations had lower abundances of species with higher SLA and greater maximum height, wood density and seed size, with seed size showing the strongest interaction with elevation. Only seed size-elevation interaction differed between fragments and contiguous forest; fragments had larger-seeded species at higher elevations (Figure 1).
In contiguous forest, lower water deficit (less negative CWD) corresponded with increased abundance of shorter statured and smaller-seeded species, and these relationships did not change for fragments (Table 1). Although SLA did not interact with any environmental gradient to shape abundances in contiguous forest, in fragments, higher SLA species increased as water deficit decreased (lower CWD) and declined with higher values of climate axis 1, i.e., wetter, cooler sites. Larger values of climate axis 1 also associated with larger-seeded species in contiguous forest, with no change in fragments. However, in fragments, larger-seeded species declined with larger values of climate axis 2 (warmer sites) even though climate axis 2 did not influence trait-mediated abundances in contiguous forest. Trends remained similar with PCA axes, which explained similar variation in the data as individual traits (Table S1). Overall, trait-environment interactions at most explained ~16% of the variation in species abundances, with maximum height being the trait that best explained change in species abundance across climate gradients (Table 1). Except for the model with maximum height, random effects explained more variation than fixed effects (Table 1, see conditional vs. marginal R2).
RLQ and fourth-corner analysis: Only contiguous forest showed significant trait-environment linkages in driving compositional change—the first RLQ axis explained 76.4% and second axis 21.1% of the cross-variance in traits and environment (Figure 2a). In fragments, 88% of the cross-variance between traits and environment was explained by the model (Figure 2b), but results were not statistically significant after accounting for multiple comparisons. In contiguous forest, fourth-corner analysis revealed that trait-mediated structuring of communities across environmental gradients occurred along multiple pathways (Figure 3a). Trait axis R1 showed significant positive correlation with environmental axis Q1, which represented a trait syndrome defined by a combination of wood density, SLA and seed size. Correspondingly, sites defined by a combination of greater elevation, higher C:N ratio, and less negative CWD had significantly higher SLA and seed sizes (Figure 3b). Elevation showed a positive correlation with the syndrome represented by trait axis 1 representing phenotypes of smaller seeded, shorter, and thicker-leaved species (Figure 3c). Fourth corner analysis for fragments revealed no significant correlations between trait axes and broad-scale environmental axes, when corrected for multiple comparisons. Moreover, no individual trait showed significant correlation with any individual environmental variable in either fragments or contiguous forest.
Trait covariance: All coefficients for the models of trait covariance are provided in Table S3. In fragments, WD-SS covariance showed a significant negative correlation with CWD, while this relationship was not statistically significant in contiguous forest (Figure 4a). SS-MH covariance decreased in wetter, warmer sites (higher values of climate axes 1 and 2), and increased with greater C:N ratio in contiguous forest (Figure 4b, 4c). In fragments, the relationship with both climate axes reversed. SS-MH covariance increased with greater variance in seed size in contiguous forest, with no significant change in this relationship within fragments. In contiguous forest, SLA-WD covariance decreased with declining climatic water deficit and increasing soil CN ratio.
SLA-WD covariance had positive and negative correlations respectively with multi-trait functional dispersion (FDis) and variance in SLA, both correlations changed sign in fragments (Figure 5, a-c). SLA-MH covariance did not correlate with any predictor. SLA-SS covariance decreased with larger variance in SLA in contiguous forest, and this remained similar in fragments (Figure 5d). WD-MH covariance in contiguous forest was only correlated with biotic variables: decreasing with greater FDis and increasing with greater variance of both traits (Figure 5, g-i). Only the correlation with MH variance changed in fragments, becoming strongly negative. Unlike for trait-mediated abundances, elevation alone did not explain trait covariance as well as multiple climate variables together (Table S4).