Fig. 5: Predictors of plant coexistence according to random forest.
Results from a random forest classifier using each variable (including maximum net PSF value, Imax) to predict whether all five species coexist until the end of the simulation. a Importance of each variable in predicting plant diversity maintenance, where importance is the mean decrease in the model’s classification accuracy when a variable is permuted (normalized by the standard deviation). b Conditional feature contribution of PSF strength to plant diversity. Each point represents the relative contribution of Imax to the predicted value of plant diversity of an independent model run. A model contribution value >0 means that the feedback value for that run contributes positively to the prediction of species coexistence. Point colours indicate the proportion of heterospecific neighbours for each tree at the end of the run. A value of zero represents a forest dominated by one species (100% conspecific neighbours), and a value of 1 means that no two adjacent plants are of the same species. Panel (b) shows that negative feedback maintains diversity, and is associated with spatial dispersion of plants. Strong positive feedback can also maintain diversity by reducing species turnover, but this represents an extreme case unlikely to be stable through time. See Supplementary Fig. 2 for the conditional feature contributions of every variable.
