Developing dynamic mechanistic species distribution models: predicting bird-mediated spread of invasive plants across northeastern North America.

Species distribution models are a fundamental tool in ecology, conservation biology, and biogeography and typically identify potential species distributions using static phenomenological models. We demonstrate the importance of complementing these popular models with spatially explicit, dynamic mechanistic models that link potential and realized distributions. We develop general grid-based, pattern-oriented spread models incorporating three mechanisms--plant population growth, local dispersal, and long-distance dispersal--to predict broadscale spread patterns in heterogeneous landscapes. We use the model to examine the spread of the invasive Celastrus orbiculatus (Oriental bittersweet) by Sturnus vulgaris (European starling) across northeastern North America. We find excellent quantitative agreement with historical spread records over the last century that are critically linked to the geometry of heterogeneous landscapes and each of the explanatory mechanisms considered. Spread of bittersweet before 1960 was primarily driven by high growth rates in developed and agricultural landscapes, while subsequent spread was mediated by expansion into deciduous and coniferous forests. Large, continuous patches of coniferous forests may substantially impede invasion. The success of C. orbiculatus and its potential mutualism with S. vulgaris suggest troubling predictions for the spread of other invasive, fleshy-fruited plant species across northeastern North America.

Multivariate forecasts of potential distributions of invasive plant species.

The fact that plant invasions are an ongoing process makes generalizations of invasive spread extraordinarily challenging. This is particularly true given the idiosyncratic nature of invasions, in which both historical and local conditions affect establishment success and hinder our ability to generate guidelines for early detection and eradication of invasive species. To overcome these limitations we have implemented a comprehensive approach that examines plant invasions at three spatial scales: regional, landscape, and local levels. At each scale, in combination with the others, we have evaluated the role of key environmental variables such as climate, landscape structure, habitat type, and canopy closure in the spread of three commonly found invasive woody plant species in New England, Berberis thunbergii, Celastrus orbiculatus, and Euonymus alatus. We developed a spatially explicit hierarchical Bayesian model that allowed us to take into account the ongoing nature of the spread of invasive species and to incorporate presence/absence data from the species' native ranges as well as from the invaded regions. Comparisons between predictions from climate-only models with those from the multiscale forecasts emphasize the importance of including landscape structure in our models of invasive species' potential distributions. In addition, predictions generated using only native range data performed substantially worse than those that incorporated data from the target range. This points out important limitations in extrapolating distributional ranges from one region to another.