Modeling the distribution of an invasive riparian plant species: Understanding drivers of invasion and testing model assumptions

Background/Question/Methods

Invasive species cause substantial economic and ecological harm throughout the world. Species distribution models are one tool used to combat invaders by providing information about areas susceptible to invasion and highlighting anthropogenic factors that promote establishment. An assumption of distribution models is that modeled species are at equilibrium with their environment. This may frequently be violated by invasive species because they have had limited time to disperse to all suitable habitat within their introduced range. We compared distribution models of the aggressive invasive riparian species Phalaris arundinacea and the functionally similar native riparian species Calamagrostis canadensis to evaluate whether the invasive species appears to violate the assumption of equilibrium and to better understand drivers of Phalaris invasion. We added a dispersal parameter to some models to account for the possibility that species have not yet reached equilibrium. We predicted that (1) models without a dispersal parameter would fit the native species better than the invasive species and (2) the addition of a dispersal parameter would improve model fit for the invasive species more than fit of the native species. We modeled distribution using logistic regression, classification trees, random forests, and boosted trees and compared model results using two accuracy metrics.

Results/Conclusions

Final explanatory variables selected and the direction of the relationship between the predictors and species’ distributions were fairly consistent between models. Phalaris is associated with sites that are warmer, drier in winter months, wetter in summer months, closer to roads, and ungrazed. Calamagrostis is found at sites that are cool, wet, more isolated from roads, and ungrazed. Both our predictions were invalidated: (1) models without the dispersal parameter performed similarly for both species with both accuracy measures, and (2) incorporation of the dispersal parameter improved model fit for both species. However, model fit for the native species improved more than model fit for the invasive species. There are several possible explanations for this finding: (a) phenomena other than equilibrium may control the ability for models to fit species; (b) Phalaris could actually be closer to equilibrium than the native species because it may be less dispersal limited; or (c) the dispersal parameter used in recent distribution models including this study may not successfully capture dispersal processes. Overall, model results provide a better understanding of Phalaris invasion as well as potential limitations in species distribution models.