Crowd-sourced trait data can be used to delimit global biomes

Biomes and their biogeographic patterns have been derived from a large variety of variables including species distributions, bioclimate or remote sensing products. Yet, whether plant trait data are suitable for biome classification has rarely been tested. Here, we aimed to assess systematically which traits are most suitable for biome classification. We derived patterns of 33 different traits by combining crowd-sourced species distribution data and trait data from the TRY database. Using supervised cluster analyses, we developed biome classification schemes using these traits and 31 different biome maps. A sensitivity analysis with randomly sampled combinations of traits was performed to identify traits and biome maps that are most appropriate for biome classification and achieved the highest data-model agreement. Due to gaps in the trait data, species distribution models were used to obtain biome maps at the global scale. We showed that traits can be used for biome classification and that the most appropriate traits are conduit density, rooting depth, height, and different leaf traits, including specific leaf area and leaf nitrogen. Data-model agreement was maximized when biome maps used to inform cluster analyses were based on biogeographic zonation and species distributions, in contrast to biome maps derived from optical reflectance. The availability of crowd-sourced trait data is heterogeneous and large data gaps are prevalent. Nonetheless, it is possible to derive biome classification schemes from these data to predict global biome patterns with good agreement. Filling data gaps is essential to further improve trait-based biome maps.