Data from: Multiple environmental drivers structure plant traits at the community level in a pyrogenic ecosystem|生态学数据集|植物特征数据集

Trait-based approaches offer a way to predict changes in community structure along environmental gradients using measurable properties of individuals. Promoted as being generalizable across systems, trait-based approaches benefit from information about the environmental drivers of trait variation, how they interact and how they change with scale. However, for most diverse, natural communities, it is largely unknown whether the relationships between leaf-level traits and interacting environmental drivers (e.g. fire, water availability) are influenced by the scale of trait aggregation. We show that landscape-level differences in community composition in a diverse, fire-dependent pine savanna are explained by a small subset of species groups that are strongly correlated with soil moisture and elevation, but are insensitive to the time since the last fire. We used a trait-based approach to show that significant variation in the community-weighted mean (CWM) of specific leaf area (SLA) and leaf dry matter content (LDMC), two traits known to drive community structure and function, was explained by a small set of factors including the time since the last fire, soil moisture, precipitation and Shannon diversity.
We show that statistical inference about the environmental drivers of community traits is radically altered when using CWMs computed with landscape-level rather than plot-level means, even over modest spatial scales. Synthesis: Environmental drivers of community composition across the landscape differed from those explaining trait composition. CWM traits were strongly influenced by interactions between drivers. Fire, in particular, strongly mediated the effect of other environmental variables on LDMC, showing that strong environmental gradients cannot be considered independently when assessing their effects on functional traits. The importance of environmental variables such as fire was lost when using landscape-level trait means, highlighting the importance of local trait variation. This suggests caution when using traits from distant populations to make inferences about local processes, especially across strong gradients.