摘要
新型统计方法和多源、多尺度空间信息数据的产生促进了物种空间分布模型的快速发展。不同的物种空间分布模型在生态学理论的运用以及前提假设上存在差异。选用不同的模型方法和输入数据会带来预测结果的不确定性。对比并集成多个物种空间分布模型,同时利用多组输入数据可降低预测的不确定性,提高物种分布模拟的精度。本文以中国特有种铁杉(Tsuga chinensis)为例,运用基于R语言开发的BioMod软件包对比9个物种空间分布模型对铁杉的模拟效果。最后以曲线下面积(ROC)为权重集成9个模型的模拟结果,产生和筛选最佳的铁杉潜在空间分布图。研究发现随机森林模型(RF)的模拟效果最好,其次是多元适应回归样条函数模型(MARS)和广义相加模型(GAM),模拟效果最差的是表面分布区分室模型(SRE)。模型集成结果显示,最适宜铁杉分布的区域集中在中国的西南及四川盆地周围,其次零星分散于华南和台湾部分地区。这一结果与前人对铁杉自然分布的描述和研究结果较为吻合。研究进一步表明,通过模型的集成能有效地降低由于单个模型所带来的模拟结果不确定性,从而提高模拟的精度和效果。
The integration of new statistical techniques and increasing availability of multi-sources and muhi-scale data sets promote the development of species distribution modeling. Yet, choice of data sets, different model types and their underlying ecological theories and assumptions can cause uncertainty in model predictions. In order to de- crease prediction uncertainty, studies using model ensemble are gaining in popularity. In this paper we apply the Bi- oMod package developed under R environment to predict the spatial distribution of Tsuga chinensis using nine different models. Our aims were to evaluate model performance, select explanatory variables, and assemble the best pre- dictive output. Random Forest, MARS and GAM performed the best amongst the nine models compared, while SRE was the worst. The ensemble models predicted that the areas of high probability for T. chinensis presence lie mainly in Southwest China and the periphery of the Sichuan basin, and are also distributed sporadically in South China and Tai- wan. These predictions reflect the actual distribution pattern of T. chinensis, and show high agreement with other ana- lyses. The application of BioMod for model ensemble lowers uncertainty and improves the prediction performance.
出处
《植物分类与资源学报》
CSCD
北大核心
2013年第5期647-655,共9页
Plant Diversity
基金
中国科学院知识创新工程重要方向项目--西南野生生物资源的挖掘与利用(KSCX2-EW-J-24)
