The effectiveness of a single regional model in predicting non-native woody plant naturalization in five areas within the Upper Midwest (United States)

Abstract

Numerous predictive models have been developed to determine the likelihood that non-native plants will escape from cultivation and potentially become invasive. Given the substantial biological and economic costs that can result from the introduction of a new invasive plant and the unending pressures of world trade and transport, the creation and implementation of effective predictive models are becoming increasingly important. One key question in the development of such models focuses on the geographic scope at which models can best be developed and applied. We have developed models to predict woody-plant naturalization in five local areas within the Upper Midwest (United States). Herein, we consider whether naturalization can be reasonably predicted from a single model for the entire region or whether local models are required for each specific area. We develop a random forest model to predict the probability of naturalization in the region and compare out-of-sample prediction errors between the regional and local models. The regional model makes better predictions of the probability of naturalization for those species observed to naturalize but worse predictions for those not currently observed to naturalize. This model development process has given us an opportunity (not previously addressed in the literature) to examine the strengths and weaknesses of local and regional approaches, with the ultimate intent of optimizing geographic scope.