Challenges in predicting the future distributions of invasive plant species

Species distribution models (SDMs) are increasingly used to predict distributions of invasive species. If successful, these models can help managers target limited resources for monitoring and controlling invasive species to areas of high invasion risk. Model accuracy is usually determined using current species distributions, but because invasive species are not at equilibrium with the environment, high current accuracy may not indicate high future accuracy. I used 1982 species distribution data from Bolleswood Natural Area, Connecticut, USA, to create SDMs for two forest invaders, Celastrus orbiculatus and Rosa multiflora. I then used more recent data, from 1992 and 2002, as validation data sets to determine how model accuracy changed over time and if current and future accuracy were related. I also tested if three alternative approaches - iterative modeling, alternative methods of choosing suitability thresholds and using a risk assessment framework - improved accuracy in predicting future distributions. Model accuracy declined over time with greater declines for models of the species (Celastrus) with the higher initial accuracy. By 2002, 49% of Celastrus and 85% of Rosa new occurrences were correctly predicted by models. Neither iterative modeling nor alternative thresholds improved accuracy of predicting 2002 occurrences, but a risk assessment framework showed promise for guiding monitoring efforts. These results suggest that measures of current accuracy may not indicate a model's predictive accuracy and must be used cautiously. Distinguishing between predictions of current and future distributions is critical. While iterative models were not successful in this study, I argue that using models in a risk assessment framework closely tied to monitoring will greatly increase the utility of SDMs for managing invasive species. (C) 2012 Elsevier B.V. All rights reserved.