关键词: 小兴安岭, Logistic回归, 环境因子, 敏感性, 针叶树种

Abstract:

Aims Our objective is to predict the distribution of coniferous species and determine species' sensitivity to environmental factors in the Small Xing'an Mountains, China. This information is important for forest regeneration and biodiversity conservation and is useful to policy makers for forest management at different scales.
Methods Slope, aspect, compound topographic index, elevation, topographic position index and average annual temperature and precipitation were selected for use in Logistic regression to predict the occurrence of Korean pine (Pinus koraiensis), larch (Larix gmelinii), fir (Abies nephrolepis), spruces (Picea koraiensis and P. jezoensis) and Mongol scotch pine's (Pinus sylvestris var. mongolica). Relative operating characteristic (ROC) was used to evaluate the Logistic models. Because Logistic regression is difficult to explain, we used the correlation coefficient between species distribution and environmental factors to explore the sensitivity of species to environmental factors.
Important findings The fir model had best fitness (ROC >80%), and Korean pine, spruces and larch had good fitness (ROC = 60%-80%). The predictions of Mongol scotch pine, a rare species, had low accuracy. At the regional scale, Korean pine, fir, spruces and Mongol scotch pine were the most sensitive to average annual precipitation, and larch was the most sensitive to slope. At local scale, Korean pine was the most sensitive to slope, fir and spruces to elevation and larch to slope position. In addition, this shift of the most important factor in species occurrence at different scale was accompanied by a change of distribution type. Korean pine was more clustered at regional scale (ROC=78.6%) than at local scale (74.4%), and its proportion increased in Fenglin Natural Reserve. Fir and spruces clustered at regional scale, but approached random distribution at local scale (ROC<60%), i.e., they occurred everywhere. Larch, the dominant species in the north, became more clustered at local scale (ROC increased from 71.7 to 82.0%) and was restricted to sites such as valleys. Scaling down from regional to local scale resulted in the changes of the most important factors affecting species distribution. Most species were sensitive to climatic factors at regional scale and geographic factors at local scale; however, larch was always the most sensitive to geographic factors due to the temperature inversion in the region.

Key words: Small Xing'an Mountains, Logistic regression, environmental factors, sensitivity, coniferous trees