Comparing predicted distribution models and hotspots analysis: A case study in Taiwan

   Describing and explaining temporal and spatial distribution patterns of species are important topics in conservation biology and biogeography. Besides, due to human impact, many biodiversity hotspots are subjected to destruction and loss. Therefore, to locate biodiversity hotspots as early as possible and to develop adequate conservation methods are of great urgency. Through finer resolution (1*1km), we used the endemic bird species in Taiwan as examples to determine: (1) the potential distribution patterns of 14 endemic species, except for Pycnonotus taivanus; (2) the applicability of four potential distribution predicted models (logistic regression (LR), multiple discriminant analysis (MDA), genetic algorithm for rule-set prediction (GARP) and artificial neuron networks (ANN) ); (3) the locations of biodiversity hotspots of the endemic species and the overlap between hotspots and protected areas. The results showed that according to visible environmental factors, GARP and LR were the best distribution predicted models whether for different individual species or classification species. Further, these results allowed nonlinear models were prior to linear models and the information about species presence was relatively important in Taiwan. The environmental factors differentiated the species presence and we demonstrated that the predicted potential distribution patterns of the endemic species in Taiwan were similar to actual distribution patterns. Finally, most biodiversity hotspots of the endemic bird species in Taiwan were protected by national parks, almost 25%, especially Shei-Pa National Park and Yu-Shan National Park. In conclusion, through a series of research of distribution predicted models, we indeed can identify the important relations between environmental features and species presence and apply these results to adaptive conservation management.
Key-words: potential distribution, models, biodiversity hotspots, logistic regression, multiple discriminant analysis, GARP, artificial neuron networks, conservation biology, biogeography