Data from: Integrating correlation between traits improves spatial predictions of plant functional composition

Functional trait composition is increasingly recognized as key to better understand and predict community responses to environmental gradients. Predictive approaches traditionally model the weighted mean trait values of communities (CWMs) as a function of environmental gradients. However, most approaches treat traits as independent regardless of known trade-offs between them, which could lead to spurious predictions. To address this issue, we suggest jointly modeling a suit of functional traits along environmental gradients while accounting for relationships between traits. We use generalized additive mixed effect models to predict the functional composition of alpine grasslands in the Guisane Valley (France). We demonstrate that, compared to traditional approaches, joint trait models explain considerable amounts of variation in CWMs, yield less uncertainty in trait CWM predictions and provide more realistic spatial projections when extrapolating to novel environmental conditions. Modeling traits and their co-variation jointly is an alternative and superior approach to predicting traits independently. Additionally, compared to a “predict first, assemble later” approach that estimates trait CWMs post hoc based on stacked species distribution models, our “assemble first, predict later” approach directly models trait-responses along environmental gradients, and does not require data and models on species’ distributions, but only mean functional trait values per community plot. This highlights the great potential of joint trait modeling approaches in large-scale mapping applications, such as spatial projections of the functional composition of vegetation and associated ecosystem services as a response to contemporary global change