Rare plant to pest plant: Can traits predict where vascular species fall along this continuum?

Background/Question/Methods

Central to the goal of preserving biodiversity is the ability to classify species. For instance,  one might like to predict which species are likely to become invasive once introduced to new regions. Alternatively, forecasting  which species are likely to be threatened with extinction under projected regimes of anthropogenic change could be used to prioritize species for intensive management.   Accordingly, whether either class of “at risk” species can be identified on the basis of biotic traits, and to what degree the two classes form extremes of a trait continuum, are important questions for basic conservation research.

We tested whether rare (n=1,925 from ranks assigned by the Nature Conservancy) or invasive (n=284 based on Plants National Database watch lists) species can be distinguished from the set of all vascular plant species native to the continental U.S. and Canada (n=13,185) on the basis of biotic traits and phylogeny.  We compiled data for species on seed mass, chromosome number, pollination syndrome, wetland habitat affinities, environmental tolerances, and phylogenetic relationships from large on-line databases.  Boosted regression trees, which allowed for non-linear responses, complex interactions, and missing data, were used to build classification models which were then evaluated using withheld test data. 

Results/Conclusions

Performance, measured as area under the ROC curve (AUC) was moderate (AUC=0.76) in predicting rare species and high (AUC=0.91) in predicting invasive species.  In terms of trait patterns, rare species were more likely to have specialized habitat preferences, small numbers of chromosomes, very large or very small seeds, and insect-pollination, whereas invasive species were characterized by large seeds relative to congeners,  medium absolute seed mass, and generalist habitat preferences.   Both rare and invasive species were more likely to have extremely high or low chromosome numbers relative to congeners.  Rarity, but not invasiveness, showed a strong phylogenetic signature. 

Thus, rather than representing opposite ends of extremes, rare and invasive species show distinct patterns in their relationships to traits.  A key finding from these results  that predicting classes of rare or invasive species appears tractable, provided adequate trait data is available, and that prediction is greatly facilitated by machine learning algorithms.