Predict first–assemble later versus assemble first–predict later: Revisiting the dilemma for functional biogeography

Abstract Predicting community‐level trait indices, such as community‐weighted mean or functional dispersion, in space time, is crucial, they are markers of ecosystem functions, proxies for services, and now integral part conservation planning. On one hand, since members a species share similar traits, natural way to model distributions function the environment first (i.e. with distribution models), then reconstruct indices any locality interest ( predict‐first ). other can be seen direct result environmental filtering, thus their may directly modelled response assemble‐first Although relatively different, these two approaches have been used interchangeably trait‐based ecology unknown consequences on usability. Here, using plant community (4463 plots) data (LNC, leaf nitrogen content; PLH, height; SLA, specific area, >800 species) covering French Alps, we compared predict dispersion accounting not abundance (CM, mean; CWM, weighted FDis, dispersion; uFDis, unweighted dispersion). We tested both interpolation versus extrapolation capabilities approaches. To support empirical findings, also run same comparative analysis simulated data. While produced skillful CM/CWM predictions R 2 > 0.63 CWM PLH) interpolation, outperformed approach unobserved conditions = 0.60 against 0.54 PLH). Functional was generally less well predicted, although again provided better 0.31 0.27 FDis 0.28 0.18 The systematically overestimated FDis. Results were confirmed by simulation experiments. conclude that modelling provides more robust spatial over time should preferred. This does suffer from having too many predicted approach.