PARENT SESSION
Monday, August 7, 5:00-6:30 pm
Poster Session 2 - Invasive species
Exhibit Hall, Ballroom Level, Cook Convention Center

Comparing presence-only models with logistic regression in predicting the potential distributions of invasive species.

Jones, Chad^*,1, Acker, Steven², Halpern, Charles¹, ¹ University of Washington, Seattle, WA² National Park Service, Port Angeles, WA

ABSTRACT- Species distribution models are increasingly used to predict the potential distributions of invasive species. These models can be used to focus monitoring and control efforts to high-risk areas. However, the majority of models for exotic species have used logistic regression, which suffers from several serious drawbacks. First, these models may perform poorly for species that are expanding their ranges because absences can also reflect lack of dispersal. Second, logistic regression is sensitive to co-linearity among habitat variables. Finally, logistic regression models are sensitive to the prevalence of the species being modeled. New modeling techniques based only on species presence, such as Ecological Niche Factor Analysis (ENFA), overcome some of these issues and may be more suitable for modeling exotic species. We compared the ability of logistic regression and ENFA to model the potential distribution of five exotic species on the Olympic Peninsula, WA. We partitioned the species data, and ran models for each subset with each technique. We then compared consistency, sensitivity to variable selection and species prevalence, and predictive success for each model type. Logistic regression models were more variable than ENFA and were more sensitive to variable selection and species prevalence. However, while logistic regression models had slightly more false positives they had many fewer false negatives leading to higher overall correct classification rate and Kappa statistics. These results suggest that although presence-only techniques such as ENFA resolve some of the concerns with logistic regression, they do not necessarily lead to increased prediction success.

Key words: invasive species, species distribution models, temperate coniferous forest