Invasive species that exhibit different distributional patterns in their native and invasive ranges create challenges for predictive modeling of invasive species spread. Caution has been raised about niche shifts in invasive species distribution modeling, but they are often ignored and have been estimated in only a few cases. Our study quantifies the magnitude of niche shifts for a suite of species and the frequency of niche shifts during the invasion process across species. We constructed spatially-explicit Bayesian logistic regression models to compare distribution drivers for widespread invasive New England plant species that are native to Japan. Unlike many species distribution models, both climate and land-use-land-cover (LULC) variables were included for both ranges because both strongly influence where species grow. We used two methods to evaluate niche shifts: 1) comparisons of standardized regression coefficients across ranges and 2) reciprocal predictions of models calibrated in one range and then used to predict the species’ distribution in the opposite range. If niches are conserved during the invasion process, we expect models calibrated in each range to have similar climatic regression coefficients. Expectations are different for LULC drivers, such that LULC categories that facilitate spread should be more important in the invasive range.
Results/Conclusions
Results reveal that distributional drivers are quite different in New England and Japan for four of New England’s most widespread invasive plants and that some species have larger shifts than others. For example, Rosa multiflora (multiflora rose) is driven more by temperature than precipitation in New England relative to Japan, where species occurrence is lower in more seasonally wet areas. Conversely, precipitation influences Celastrus orbiculatus (Oriental bittersweet) occurrence more than temperature in both ranges, but the species occurs in areas with less seasonal temperature variation in Japan relative to New England. The inclusion of LULC substantially improved model performance across species and in both ranges and demonstrated the importance of LULC categories facilitating species movement in the invasive range. The comparison of reciprocal predictions for each species confirmed that prediction accuracy depends on the similarity of fitted regression coefficients across ranges. Given the prevalence of niche shifts among invasive species, our study shows the value of combining native and invasive range data to generate more reliable predictions of invasive species distributions.