Influence of nonclimatic factors on the habitat prediction of tree species and an assessment of the impact of climate change

To determine the influence of nonclimatic factors on predicting the habitats of tree species and an assessment of climate change impacts over a broad geographical extent at about 1 km resolution, we investigated the predictive performance for models with climatic factors only (C-models) and models with climatic and nonclimatic factors (CN-models) using seven tree species in Japan that exhibit different ecological characteristics such as habitat preference and successional traits. Using a generalized additive model, the prediction performance was compared by prediction accuracy [area under the operating characteristic curve (AUC)], goodness of fit, and potential habitat maps. The results showed that the CN-models had higher predictive accuracy, higher goodness of fit, smaller empty habitats, and more finely defined borders of potential habitat than those of the C-models for all seven species. The degree of the total contribution of the nonclimatic variables to prediction performance also varied among the seven species. These results suggest that nonclimatic factors also play an important role in predicting species occurrence when measured to this extent and resolution, that the magnitude of model improvement is larger for species with specific habitat preferences, and that the C-models cannot predict the land-related habitats that exist for almost all species. Climate change impacts were overestimated by C-models for all species. Therefore, C-model outcomes may lead to locally ambiguous assessment of the impact of climate change on species distribution. CN-models provide a more accurate and detailed assessment for conservation planning.