Species clustering, goodness-of-fit and predictive ability of the whole hierarchical trees of species clustering.

<p>(A), (C) and (E) Microbial experiment of Langenheder et al. [<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0201135#pone.0201135.ref016" target="_blank">16</a>]. (B), (D) and (F) Biodiversity II experiment of Tilman et al. [<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0201135#pone.0201135.ref017" target="_blank">17</a>]. (A) and (B) Modelling of ecosystem function based on the hierarchical tree of species clustering, from 1 to 4 functional groups of species in both experiments (see <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0201135#pone.0201135.g002" target="_blank">Fig 2E and 2F</a>). Each bar corresponds to the error induced by leaving out the ecosystem to predict. Solid red line is the 1:1 line. (C) and (D) Hierarchical tree of species clustering. Dotted red line corresponds to the optimum number of functional groups indicated by <i>AICc</i>. Functional groups are noted by italic lowercase letters. (E) and (F) Boxplots of ecosystem functions by assembly motifs.