期刊
ECOLOGICAL MODELLING
卷 246, 期 -, 页码 79-85出版社
ELSEVIER
DOI: 10.1016/j.ecolmodel.2012.07.018
关键词
Climate envelope model; Endangered species; Species distribution model
类别
资金
- U.S. Fish and Wildlife Service
- National Park Service (Everglades and Dry Tortugas National Park through the South Florida and Caribbean Cooperative Ecosystem Studies Unit)
- U.S. Geological Survey (Greater Everglades Priority Ecosystems Science)
Climate envelope models are widely used to forecast potential effects of climate change on species distributions. A key issue in climate envelope modeling is the selection of predictor variables that most directly influence species. To determine whether model performance and spatial predictions were related to the selection of predictor variables, we compared models using bioclimate variables with models constructed from monthly climate data for twelve terrestrial vertebrate species in the southeastern USA using two different algorithms (random forests or generalized linear models), and two model selection techniques (using uncorrelated predictors or a subset of user-defined biologically relevant predictor variables). There were no differences in performance between models created with bioclimate or monthly variables, but one metric of model performance was significantly greater using the random forest algorithm compared with generalized linear models. Spatial predictions between maps using bioclimate and monthly variables were very consistent using the random forest algorithm with uncorrelated predictors, whereas we observed greater variability in predictions using generalized linear models. (C) 2012 Elsevier B.V. All rights reserved.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据