Predicting the potential distribution of an invasive plant using stratified sampling and habitat modeling.

摘要

Habitat models used to predict the distribution of invasive species are often produced using compiled presence data collected for purposes other than modeling. These localities are likely to be autocorrelated and may create biased models. To address potential bias, I used a stratified scheme to collect presence data. I sampled for Tibouchina herbacea, an invasive herb from South America, across stratified zones on Kohala Mountain, Hawai'i, based on four environmental gradients: vegetation, slope, rainfall and elevation. The sampled localities represented the diversity of environments that characterize T. herbacea's current distribution, and were used to create a predictive habitat model with the GARP genetic algorithm. The model had 7% omission error, and over-predicted presence, with a 35% commission error. Over-predicted locations were likely sites of further range expansion of this invasive species. T. herbacea was associated with wet forests in both the local-scale associations as well as the habitat model predictions.