Friday, August 6, 2010

PS 93-51: Influence of non-climate factors on habitat prediction of tree species and assessing climate change impact

Motoki Higa, Yokohama National University and Nobuyuki Tanaka, Forestry and Forest Products Research Institute.

Background/Question/Methods

Many studies have attempted to predict plant habitats under current and future climates, and to assess the vulnerability of these habitats to the current-rapid climate changes. The species distribution models (SDMs) in these studies were mainly constructed by using only climatic factors, although the plant distribution range is determined not only by the climate but also by other environmental factors. However, the importance of non-climatic factors in habitat prediction was not sufficiently considered. In this study, we assessed the accuracy of two different SDMs, namely, the Climate and Climate-Land models, to reveal the influence of non-climatic variables on habitat prediction and assessment of the climate change impacts on plant distributions. Three land-generalist and three land-specialist tree species were used as the target species. The SDMs of the target species were made by generalized additive models. The distribution data of Phytosociological releve database were used as the objective variables. The C-model involved the use of only four climatic variables as the explanatory variables, and the CL-model involved the use of six non-climatic variables related to topography and geology in addition to the four climatic variables. The accuracy of both models was assessed on local and global scales by ROC analysis.

Results/Conclusions

A comparison between the two SDMs with regard to the AUC values and the predicted potential distribution maps revealed that the CL-model can predict the potential distributions of target species over more narrow areas and with higher accuracy than the C-model. Furthermore, some potential distributions were predicted only by the CL-model and were considered as land refuges under the current climate change conditions. However, the CL-model could not predict some local distributions or underestimated them, although the C-model could predict these distributions as the potential distributions. This tendency was especially apparent in the land-specialist species. It seems that the main reason for this prediction error was the geographical heterogeneity of non-climatic variables. Therefore, it is suggested that the non-climatic variables contributed considerably to the assessment of the impact of climate change on plant distributions; however, the habitats were partially underestimated when non-climatic factors were included. Hence, in order to ensure the reliable interpretation of results in this case, it will be necessary to assess the spatial relationship between the current distribution pattern of plants and the geographical heterogeneity of non-climatic variables.