Page Banner

United States Department of Agriculture

Agricultural Research Service

Title: A Test of Four Models to Predict the Risk of Naturalization of Non-native Woody Plants in the Chicago Region

Authors
item Widrlechner, Mark
item Thompson, Janette -
item Kapler, Emily -
item Kordecki, Kristen -
item Dixon, Philip -
item Gates, Galen -

Submitted to: Journal of Environmental Horticulture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: August 14, 2009
Publication Date: December 1, 2009
Citation: Widrlechner, M.P., Thompson, J.R., Kapler, E.J., Kordecki, K., Dixon, P.M., Gates, G. 2009. A Test of Four Models to Predict the Risk of Naturalization of Non-native Woody Plants in the Chicago Region. Journal of Environmental Horticulture. 27(4):241-250.

Interpretive Summary: Nursery and landscape professionals continue to search for new woody landscape plants that provide consumers with visual interest and diversity in a variety of managed settings. Expanding the palette of both native and non-native species used in the landscape is of concern to horticulturists, scientists, managers of natural areas, and some members of the general public due to the potential for widely used introductions to become invasive pests. Although relatively few non-native woody species naturalize, and even fewer are likely to become invasive, invasions can have significant impacts, and their control is difficult and costly. One alternative to lengthy test periods before new woody plants are released for introduction is the use of predictive modeling to assess risk associated with certain plants. Reliable models that predict plant invasiveness can allow time-consuming and expensive field screening to be focused only on species of greatest concern or those for which basic information is lacking. This paper describes efforts to find accurate, rapid, and relatively easy-to-use methods for identifying species that could become invasive pests by testing four models to predict naturalization (first tested in Iowa) on two new sets of data for non-native woody plants cultivated in the Chicago region. None of the four models has an optimal combination of power (classification rates) and accuracy (error rates) to support their unmodified use by nursery professionals, regulatory officials, or land managers. A combination of approaches to capture both high classification rates and low error rates will likely be the most effective until improved protocols based on multiple regional datasets can be developed and validated. Efforts are now underway to evaluate new datasets from the Midwestern U.S. to develop and test new protocols.

Technical Abstract: Accurate methods to predict the naturalization of non-native woody plants are key components of risk-management programs being considered by nursery and landscape professionals. The objective of this study was to evaluate four decision-tree models to predict naturalization, first tested in Iowa, on two new sets of data for non-native woody plants cultivated in the Chicago region. We identified life-history traits and native ranges for 193 species (52 known to naturalize and 141 not known to naturalize) in two study areas within the Chicago region. We used these datasets to test four models (one continental-scale and three regional-scale) as a form of external validation. Application of the continental-scale model resulted in classification rates of 72-76%, horticulturally limiting (false positive) error rates of 20-24%, and biologically significant (false negative) error rates of 5-6%. Two regional modifications to the continental model gave increased classification rates (85-93%) and generally lower horticulturally limiting error rates (16-22%), but similar biologically significant error rates (5-8%). A simpler method, the CART model developed from the Iowa data, resulted in lower classification rates (70-72%) and higher biologically significant error rates (8-10%), but, to its credit, it also had much lower horticulturally limiting error rates (5-10%). A combination of models to capture both high classification rates and low error rates will likely be the most effective until improved protocols based on multiple regional datasets can be developed and validated.

Last Modified: 11/26/2014
Footer Content Back to Top of Page