随机森林算法在新疆物种丰富度影响因素研究中的应用

基于多源遥感数据及其产品，以2010年新疆地区鸟类与哺乳动物物种丰富度空间分布数据为基础，结合土地利用、植被、气候、地形等遥感数据产品，探讨了影响新疆地区鸟类和哺乳动物物种丰富度的各环境因子空间分布及其差异.通过随机森林算法对影响鸟类和哺乳动物种数的环境解释变量进行了重要性评估，同时采用探索性回归分析对物种丰富度与环境因子的关系进行建模，比较了鸟类和哺乳动物在草地、林地和耕地三种生境中的主要环境影响因子及其差异.结果显示：影响物种丰富度的环境因子中，植被生长状态与能量转换能力对鸟类丰富度十分重要，且草地的海拔高度重要性最高，为38.38%；耕地所处的气候类型对鸟类丰富度有明显影响，其中年均温重要性达到32.98%；哺乳动物的三种生境选择中气候和海拔条件十分重要，共占重要性的60%以上，最高达76.85%，在耕地中哺乳动物更看重气候而非植被生长情况；三种生境中鸟类和哺乳动物的最优模型及其影响因子均不同，耕地对于鸟类、林地对于哺乳动物的模型解释力最高，分别为69.9%和68.9%，随机森林和探索性回归分析均显示年均温、年降水量和海拔高度是哺乳动物丰富度差异的重要解释因子.

Research on determinants of species richness in xinjiang based on random forest appraoch

Taken gridded spatial distribution data of bird and mammal species of Xinjiang in 2010 as primary data source, combined with multi-sources of remote sensing data production, such as land use, vegetation, climate, and topographic data, determinants which affected birds and mammal species richness in Xinjiang have been recognized and their spatial variation was discussed. Importance of each determinant which affected the number of bird and mammal species have been assessed and ranked using Random Forest approach. Then relationship models between species richness and determinants were built through exploratory regression analysis. Three main habitats of grassland, woodland and farmland were taken out and the difference of effect factors to birds and mammals species richness between them were compared and analyzed. The results show that among all the determinants, vegetation growing status and their energy conversion capability were the most two important factors for bird richness. As for grassland habitat, the importance of altitude was 38.38%, which was highest value. Climate type in farmland had significant impacts on bird diversity, among which the importance of annual average temperature reaches 32.98%. Climate and altitude were very important considerations for mammal when choosing a habitat, accountied for more than 60% of the importance, and the highest was 76.85%. In cultivated land, climatic factors were more important than vegetation growth for mammalian habitat. Optimal models for birds and mammals under different types of habitats were varied and their variables were different accordingly. Model performance of farmland for birds and forest for mammals were the best two, which could explain the relationship between determinants and species richness about 69.9% and 68.9%, respectively. The results of both random forest and exploratory regression analysis shown that altitude, average annual temperature and precipitation were the most three important determinants to mammal species richness.