Habitat distribution models are increasingly used to predict the potential distributions of invasive species. If successful, these models can help managers target limited resources for monitoring and controlling invasive species to areas where the species are most likely to occur. However, these distribution models all share the fundamental assumption that species are at equilibrium with the environment, which is not true for most invasive species. The accuracy of these models is usually determined based on the current species distribution. However, because of the lack of equilibrium, models that accurately predict current distributions of invasive species may not be effective at predicting future distributions. I used data on the 1982 distributions of two invasive species, Celastrus orbiculatus and Rosa multiflora, in the Bolleswood Natural Area in southeastern Connecticut to create habitat distribution models using logistic regression. I then assessed how well these models predicted the distributions of the two species in 1982, 1992 and 2002, using data not included in the original model.
Results/Conclusions
Overall model accuracy (measured using AUC) for Celastrus was high using data from 1982 and 1992 but declined when using data from 2002. In addition, the ability of models to accurately predict new occurrences declined dramatically from 98% in 1982 to 76% in 1992 and only 27% in 2002. Models for Rosa followed a similar trend although predictive accuracy only dropped from 92% in 1982 to 61% in 2002. These results suggest that measures of model accuracy from the time that model is developed are not good determinants of the ability of models to predict future occurrences. Thus habitat distribution models for invasive species must be used with great care.
